Evaluation of a New Barkhausen Noise Sensor
Jesper Hamfelt 2015
Master of Science in Engineering Technology Engineering Physics and Electrical Engineering
Luleå University of Technology
Department of Computer Science, Electrical and Space Engineering
Evaluation of a New Barkhausen Noise Sensor
Jesper Hamfelt
Lule˚a University of Technology
Dept. of Computer Science, Electrical and Space Engineering
2015-03-09
A BSTRACT
In this thesis, it is investigated if the Barkhausen noise can be detected in a rotating environment using a new, passive sensor design. The sensor relies on the relative move- ment between the measured material and the sensor to function. The idea behind the thesis is to investigate if this sensor can detect changes in the Barkhausen noise mea- sured on a ferromagnetic material as it is being subjected to stress until the point of breakage, and if this change can be related to fatigue in the material. If this is the case, the sensor-technology could be valuable for e.g. condition monitoring purposes.
The work was focused on improving the design of an existing sensor, to build an accom- panying measurement system – composed of the sensor, an amplifier, a data acquisition box and a computer – and to provide a proof of concept that the sensor can detect the Barkhausen noise. The sensor in its most basic form consist of a solenoid coil, with or without a core, and a permanent magnet. A number of different sensor versions were made and tested, to investigate the effects of the following parameters: the thread size in the coil, the core material, and the number of magnets and their orientation. To test the functionality of the sensor design, a rotating bending test rig (RotaBend by Sincotec) was used. This machine also allow to expose the specimen to stress, to investigate if a change related to fatigue can be detected. MATLAB were used as an interface to the data acquisition box, to acquire and analyse data from the measurements using the sensor. The analysis were performed by calculating the RMS-voltage and the Fourier transform of the data.
The thesis shows that the sensor can detect the Barkhausen noise in a rotating element.
The limited amount of tests also indicates that it is possible to detect a change in the Barkhausen noise RMS-value as the material of the specimen is fatigued. It has not yet been proven if or how this change can be related to fatigue in the material. The tests also indicates that there is no change in the frequency content of the Barkhausen noise as the material of the specimen is fatigued. Left for further investigation is at what depth in the material the Barkhausen noise is induced.
iii
P REFACE
I would like to thank my supervisors Fredrik H¨aggstr¨om and Jonas Gustafsson for their support, encouragement and input during my work with thesis. I would also like to thank everyone at the SKF-LTU UTC for providing a stimulating work environment.
Lastly, I would like to give specially thank my loving partner, Elin, for her ever lasting support and encouragement.
Jesper Hamfelt
v
C ONTENTS
Chapter 1 – Introduction 1
1.1 Barkhausen noise . . . 2
1.2 New sensor design . . . 3
1.3 Related work . . . 4
1.4 Nomenclature . . . 6
Chapter 2 – Measurement System 7 2.1 Theory of operation . . . 8
2.2 The sensor . . . 8
2.2.1 Magnets . . . 10
2.2.2 Pick-up coil design . . . 10
2.2.3 Manufactured sensors . . . 11
2.3 Electronics . . . 12
2.3.1 Component values . . . 14
2.3.2 PCB-design . . . 15
2.4 DAQ board . . . 16
2.5 Acquisition function on the PC . . . 17
2.5.1 Implementation of AquireData . . . 19
Chapter 3 – Test Setup and Measurement Analysis 21 3.1 Rotating Bend Rig . . . 21
3.2 Signal processing and information extraction . . . 23
3.2.1 Fourier transform . . . 23
3.2.2 RMS voltage . . . 24
3.2.3 Implementation in Matlab . . . 26
Chapter 4 – Results from Measurements on the RotaBend Rig 29 4.1 Performed measurements . . . 29
4.2 Difference between AMP1 and AMP2 . . . 30
4.2.1 RMS voltages . . . 30
4.2.2 Fourier transforms . . . 31
4.3 Identifying a difference over time using the Fourier transform . . . 32
4.4 Frequency content when measuring away from the specimen or with stopped test rig . . . 34
4.5.2 7P . . . 36
4.5.3 7AXII . . . 38
4.5.4 6AX . . . 38
4.5.5 3MS1 . . . 39
4.5.6 Grad3MSI . . . 40
4.6 Comparison between frequency filtered RMS voltages . . . 41
4.6.1 0 - 1.8 kHz region RMS voltages . . . 41
4.6.2 1.8 - 3.75 kHz region RMS voltages . . . 43
4.6.3 3.75 - 4.5 kHz region RMS voltages . . . 45
4.6.4 4.5 kHz - 1 MHz region RMS voltages . . . 47
4.7 Sensor comparison . . . 50
4.8 Evaluation of speed influence on the signal . . . 52
Chapter 5 – Conclusions and Discussions 55 5.1 Use of the Fourier transform to detect fatigue . . . 55
5.2 Use of RMS voltage to detect fatigue . . . 56
5.3 Sensor and speed effect on measurements . . . 57
5.4 Proof of Concept? . . . 58
Chapter 6 – Future work 59 Appendix A – Produced plots 61 A.1 8R3, Sensor 1 Parallel, AMP2, 40Nm . . . 62
A.2 8R3, Sensor 1 Perpendicular, AMP2, 40Nm . . . 63
A.3 8R3, Sensor 1 Parallel, AMP1, 50Nm . . . 64
A.4 8R3, Sensor 1 Perpendicular, AMP1, 50Nm . . . 65
A.5 7P2, Sensor 1 Parallel, AMP1, 50Nm . . . 66
A.6 7P2, Sensor 1 Perpendicular, AMP1, 50Nm . . . 67
A.7 7P2, Sensor 2 Perpendicular, AMP1, 50Nm . . . 68
A.8 7P, Sensor 1 Perpendicular, AMP1, 50Nm . . . 69
A.9 7P, Sensor 1 Perpendicular, two, AMP1, 50Nm . . . 70
A.10 7P, Sensor 2 Parallel, AMP1, 50Nm . . . 71
A.11 7AXII, Sensor 3 Perpendicular, AMP1, 40Nm . . . 72
A.12 7AXII, Sensor 3 Perpendicular, AMP1, 50Nm . . . 73
A.13 6AX, Sensor 1 Perpendicular, AMP1, 45Nm . . . 74
A.14 3MS1, Sensor 1 Perpendicular, AMP1, 45Nm . . . 75 viii
A.15 Grad3MSI, Sensor 1 Perpendicular, AMP1, 45Nm . . . 76 A.16 Grad3MSI, Sensor 4 Perpendicular, AMP1, 45Nm . . . 77
ix
C HAPTER 1 Introduction
Condition monitoring is used to monitor parameters (vibration, temperature, sound etc.) of a machine or its components, in this case a rolling element bearing, in order to detect a change which can indicate a developing fault. The monitoring is useful to prevent critical failures, and to perform maintenance at the right time (condition based maintenance). The use of condition monitoring can reduce unnecessary waste of functioning components that have been running until their pre-calculated lifetime but have not broken, and minimize downtime due to maintenance, both of which can have economical and environmental benefits to the society. SKF has a University Technology Center (UTC) for Advanced Condition Monitoring at LTU, where the work is aimed at developing tomorrow’s technology for monitoring machines with so-called smart bearings [1]. The new sensor is intended to provide a complement to existing condition monitoring technologies, using a new approach of the existing non-destructive Barkhausen noise [2]
method. In the future it might be possible to integrate the sensor in a bearing and thus provide real time condition monitoring while the bearing is in operation.
The aim for this Master’s thesis work was to further develop the new sensor and in- vestigate if it can be used to detect change in a ferromagnetic material as it is being fatigued. A prototype of this sensor and an accompanying amplifying circuit had al- ready been produced and preliminary testing had been conducted at the UTC before this work was performed, the testing implied that the sensor could measure Barkhausen noise and therefore it was decided to start this work. The work presented in this thesis is focusing on sensor and amplifier design improvements and the challenges of making them work as a system, and to perform measurements on real steel elements as they are being subjected to stress until the point of breakage in a rotating bending test rig.
1
1.1 Barkhausen noise
The Barkhausen noise (BN) is a magnetic phenomenon named after its discoverer, the German physicist Heinrich Barkhausen [3]. He discovered it using a test setup with a coil wound around a ferromagnetic rod and connected to a speaker through an amplifier.
He then magnetized the rod by moving a magnet close to it, which induced pulses in the coil that was heard as rasping ”clicks” from the speaker. It was a very noise-like sound, hence the name ”Barkhausen noise”.
From Barkhausen’s work it has since been found that the effect is observed when a ferromagnetic material is magnetized and that it is mainly caused by magnetic domain walls making sudden jumps in orientation and thus creating small abrupt changes in the magnetic flux [4]. The amount of noise is dependent on the material parameters such as carbon contents [5] and grain size [6]. It is also affected by stress [7] and fatigue [8].
A common Barkhausen sensor design is shown in Figure 1.1. This kind of sensor is used for non-destructive testing to detect structural changes caused by stress in a measured specimen. It can also be used to detect grinding burn in the surface of the specimen [9]. To produce the BN, an alternating current is applied in the magnetizing coil, thus inducing an alternating magnetic field in the yoke and specimen, which in turn is induc- tively captured by the pick-up coil. The captured signal will include the applied field and the BN, and therefore the signal needs to be high pass filtered in order to remove the induced field from the measurement. The signal is also low pass filtered in order to avoid aliasing during digitisation and to suppress high frequent background noise [10].
For the commercial Barkhausen system Rollscan 350 [11] the inducing field is in the range of 1 - 1000 Hz, the high pass cut-off frequency in the range 10 - 200 kHz and the low pass cut-off frequency in the range 70 - 450 kHz. The use of an alternating field to create the noise puts a penetrating depth limitation on the measurement. This is due to the skin effect, which states that higher frequencies penetrate a surface less than lower frequencies [12]; and since the signal has to be high pass filtered, in order to remove the induced field, the information from deep in the material is to a great extent lost.
1.2. New sensor design 3
Pick-up
AC power supply
Yoke
Amplification and filtering Specimen
Figure 1.1: Schematic of the commonly used Barkhausen Sensor
1.2 New sensor design
To circumvent the need for high pass filtering, the proposed sensor uses the rotation of the bearing and a permanent magnet to induce the BN in the metal surface as it rotates.
The BN is recorded with a pick-up coil. The idea behind this sensor was initiated from Fredrik H¨aggstr¨om’s work on Energy Harvesting in rolling-element bearings. Figure 1.2 shows a schematic image of the new sensor measuring on a moving surface. In this case the magnet is aligned so that its flux is directed towards the surface of the specimen.
The arrows in the specimen represent the orientations of the magnetic domains.
Since the BN is created using a passive method there will be no applied frequency that needs to be attenuated, instead the only high pass filtering necessary is to suppress the 50 or 60 Hz noise arising from the mains frequency. This should make it possible to obtain information from deeper in the material than possible for the commonly used sensor, as explained earlier.
The new sensor approach also removes the need to have a connection between the sensor and the measured material, which means that it should be possible to integrate it in a bearing and use it to continuously monitor the bearings raceway during operation. This requires that it has a low power consumption, since whatever power it uses will have to be harvested inside the bearing. Due to its passive design the main power consumer will be the amplifier and it can be designed to consume very little power.
v
B
Pick-up
Amplification and filtering
Specimen Magnet
Figure 1.2: Schematic image of the proposed new Barkhausen Sensor, measuring on a specimen moving at a velocity v. The arrows in the specimen represents the orientation of the magnetic domains. The arrows from the magnet represents the orientation of the magnetic flux B.
1.3 Related work
The BN method is a type of Nondestructive Testing (NDT) method. As presented by The American Society for Nondestructive Testing [13], ”Nondestructive testing (NDT) is the process of inspecting, testing, or evaluating materials, components or assemblies for discontinuities, or differences in characteristics without destroying the serviceability of the part or system. In other words, when the inspection or test is completed the part can still be used.”. It is used in manufacturing, fabrication and in-service inspections to ensure the integrity and reliability of the tested specimen. Apart from the BN method, there are numerous of NDT methods, such as: Radiographic testing [14], a type of x-ray method; Ultrasonic Testing [14], using high-frequency sound; Visual Testing [15], using automated optics or human vision; Vibration measurements [14], analysing the vibrations of the tested specimen; Acoustic Emission [14], analysing the ”sound” the tested specimen emits; and several others. Few of these can be implemented and used during operation of a rolling bearing.
Other work in the BN area has, as stated, mostly used a setup with an external alter- nating magnetic field creating the BN. For the BN work referenced previously, [2, 6, 8], the material measured on was in a flat rectangular shape and stressed using a bending or tensing routine. The results from these papers are therefore difficult to compare with the results from this work, given the different type of sensor, the different material shape and the difference in how the stress was applied to the materials. There are however two articles which can be, and are, used to compare the results in this thesis with. The work by Palma et al. [16] which uses the same kind of test rig as in this thesis but the classic
1.3. Related work 5
sensor design, described in Section 1.1, to evaluate the BN during fatiguing of a mate- rial; and the work by Soultan et al. [17] which measures Mechanical BN during fatigue in the same kind of test rig as used in this thesis. Mechanical BN is BN that occurs without an exiting magnetic field when a material is affected by an alternating cyclic stress. Common for both articles [16, 17] is that for applied stresses over the endurance limit the BN signal increases during the early fatigue stages and then either decreases or plateaus. The measured signals varies a lot between each measurement point.
1.4 Nomenclature
Variable Explanation Unit
B Magnetic flux density T
H Magnetic field intensity A/m
Φ Magnetic flux Wb
V Voltage V
N Numbers of turns of wire in solenoid -
A Surface area of solenoid m2
µ Permeability H/m
Nd Demagnetizing factor -
Dc Solenoid core diameter m
lc Solenoid core length m
fc Cut-off frequency Hz
R Resistance Ω
C Capacitance F
Abbreviations
BN Barkhausen Noise -
C Capacitor -
DC Direct Current -
DAQ Data Acquisition -
DFT Discrete Fourier Transform -
FFT Fast Fourier Transform -
LTU Lule˚a University of Technology -
OP-Amp Operational Amplifier -
PCB Printed Circuit Board -
R Resistor -
RMS Root Mean Squared -
C HAPTER 2 Measurement System
To prove that the new sensor design can be used to detect fatigue in a material during its lifespan, the sensor and an accompanying measurement system needed to be developed.
The goal of the sensor and system design was that it should be possible to acquire the Barkhausen noise (BN) [2] and log the BN signal data on a computer for further analysis.
The system should be able to capture signals of a frequency up to 1 MHz, this was chosen in order to have a high bandwidth and to see if more information could be found well over the range of the commercial BN sensors, which measures up to 450 kHz [11].
To achieve this, a system was designed composed of the developed sensor, an amplifier, a DAQ board (Data AcQuisition) and a computer. The sensor is used to pick-up the BN, the amplifier increases the signal amplitude from the sensor so that it is large enough to sample. The sampling is performed by the DAQ board. The DAQ board is connected to the computer from which it is controlled and where the sampled data is stored. A schematic view of the system is shown in Figure 2.1. Measurements were performed on metal rods mounted in a test rig called a Rotating Bend Rig (RotaBend). The different parts are explained further in the following sections.
7
Amplifier DAQ PC
Sensor
Specimen
Figure 2.1: Schematic overview of the measurement system
2.1 Theory of operation
As stated in the introduction, the sensor is passive by design and power is only required to amplify the signal. When the sensor is placed over a moving ferromagnetic material the magnetic domains in the material will align with the external magnetic field from the magnet [4]. When the domains align they do it randomly and there sudden movements give rise to a discontinuity in the magnetic field, which is measured with the sensor. This discontinuity is the BN [2]. The amount of BN is dependent on material parameters such as carbon contents [5] and grain size [6], but is also affected by fatigue in the material [8], which is the most interesting aspect for this work.
A voltage is induced in the pick-up from flux changes parallel to the coil [18], so the constant magnetic field from the magnets will not be picked up. The BN creates a change in the field and will therefore be picked up by the sensor. Also, other disturbances such as the electromagnetic field from the mains voltage and any vibration in the measured material will be picked up. The mains frequency is attenuated by using a high pass filter with a cut-off frequency over 50 Hz, other disturbances are harder to predict and will have to be dealt with if any problem arises.
2.2 The sensor
The BN should be generated and picked up by the sensor. The design goal was not just to generate and pick up as much BN as possible but also to allow the possibility to test different sensor implementations. A way to mount the sensor when measuring was also
2.2. The sensor 9
needed.
The designed sensor to achieve this consists of a pick-up coil, with or without a fer- romagnetic core, and one or more magnets. The magnets are placed parallel on each side of the pick-up coil as shown in Figure 2.2. The pick-up is soldered to a female SMA-connector in order to more easily connect to an amplifier.
Pick-Up Coil
Magnet Optional
magnet magnet Optional Optional
Optional
Optional magnet
magnet
magnet
Core
B
Specimen
Figure 2.2: Schematic view of the sensor.
For the mounting of the sensor, a casing was designed and 3D-printed. This casing has room for a pick-up of 15 mm length and 5 mm diameter, and a total of six 5x5 mm magnets (three in each row). Figure 2.3 shows two different views of a casing, showing the two rectangular holes for the magnets and the circular for the pick-up. The SMA-connector is mounted between the ’ears’ and connected to the pick-up through a hole in the back. The whole casing is attached to a magnetic arm with a bolt through the piece sticking out from the back. Four different sensor implementations were made and used during the testing, further explained in Section 2.2.2.
Figure 2.3: Two views of the empty casing for the sensor
2.2.1 Magnets
Neodymium magnets were used for the sensors. They were mounted with their magnetic flux density, B, pointing either downwards or upwards, as indicated by the arrows in Figure 2.2. This orientation was chosen in order to see what happens when magnets are placed on each side of the pick-up, with their field densities pointing in opposite directions. The idea being that two rows of magnets with flux in opposing directions should give rise to more BN, due to the domains flipping two times in the vicinity of the pick-up.
2.2.2 Pick-up coil design
A pickup of solenoid type was used for the experiments, versions with and without a ferromagnetic core was used. The aim in the design process was to obtain a strong as possible signal. Given Faraday’s law of induction [18]
V = −NdΦ
dt = −N AdB
dt = −µN AdH
dt , (2.1)
we can see that the induced voltage V from an alternating magnetic field H = Bµ, is affected by the core material (the permeability µ = µr+ µ0), the number of turns N of the wire and the surface area of the coil A. So in order to get as much signal as possible the coil should be wound to a large area with a small diameter thread (to get more turns) around a material with a large permeability.
2.2. The sensor 11
Further, the optimal shape of a coil with a ferromagnetic core of relatively large µr is, as stated in [19], to have a large core length lc with a small core diameter Dc which minimizes the demagnetizing factor Nd, according to Equation (2.2).
Nd∼= Dc2 l2c ·
ln2lc
Dc − 1
. (2.2)
In this case, with a large µr, the resultant permeability of the core, µc, is mainly depen- dent on the demagnetizing factor Nd, as stated in Equation (2.3).
µc= µr
1 + Nd· (µr− 1), (2.3)
According to Equation (2.1) a larger permeability generates a higher voltage, and there- fore an as low as possible Ndis wanted in order to minimize its effect on the permeability.
To summarize, to pick up as much signal as possible, the pick-up coil should have a thin thread wound many times around a high permeability ferromagnetic core with a length much greater than its diameter.
2.2.3 Manufactured sensors
With the conditions explained in the previous section in mind, four sensors were pro- duced, presented in Table 2.1. The produced sensors were not all made to match the criteria for picking up the most signal. This choice, to not match the criteria, was made in order to experimentally verify the differences between optimal and non-optimal sensors. Sensor 1 and 2 can be used to test the difference between an air core and a ferromagnetic core, Sensor 2 and 3 can be used to test the influence of an additional row of magnets, thus an increased magnetic field, and Sensor 1 and Sensor 4 can be used to test the influence of a smaller diameter thread and an increased number of turns.
All pick-ups were made to have similar length and diameter, with the pick-ups for Sensor 1 and 4 having a slightly larger inner and outer diameter. The two rows of magnets in Sensor 3 were arranged with their fields pointing in opposite directions. In order to easily connect and disconnect the sensors to the amplification circuit, a female SMA-connector was soldered to each coil.
Table 2.1: Overview of the sensors used in the measurements
Sensor # Core material Thread diameter Number of magnets
1 Air 0.1 mm 3 (in one row)
2 Ferromagnetic wire 0.1 mm 3 (in one row) 3 Ferromagnetic wire 0.1 mm 6 (in two rows)
4 Air 0.05 mm 3 (in one row)
Figure 2.4 shows Sensor 1 and 3 manufactured and mounted inside the casing, with the SMA-connector soldered to the coils. The glue on the SMA-connector is there to prevent short circuiting the solenoids. The extra wires in Sensor 1 are there because the wires of the pick-up got damaged. During testing there was a piece of electrical tape glued to the front of the sensors to prevent any damage to the pick-ups.
Figure 2.4: Two of the manufactured sensors, Sensor 3 (to the left) and Sensor 1 (to the right)
2.3 Electronics
As stated earlier, the signal from the sensor needs to be amplified to a level that matches the input range of the DAQ, in order to utilise the full bit-range in the digital conver- sion. Therefore an amplifying circuit was designed and constructed. In addition to the amplification there was also a need for noise filtering, to remove unwanted noise from the mains frequency and other sources, and to prevent aliasing during digitisation. The circuit, including amplification and noise filtering should be able to amplify signals at a frequency up to 1 MHz.
2.3. Electronics 13
The designed amplifying circuit consists of two identical cascaded inverting amplifier [20]
steps. The use of two steps instead of one was done to reduce the influence of noise in the amplifier. To further reduce noise, the ultra-low noise operational amplifier (OP-amp) EL-2125, [21], was used. This OP-amp is designed to have low power consumption and a high bandwidth. Figure 2.5 shows the schematic of the amplifying section. This schematic omits the voltage supply for each step. Two decoupling capacitors are also connected close to each voltage pin on the OP-amps, as per instructed in their data-sheet:
one 100 nF ceramic and one 4.7µF tantalum.
−
+
C11
R11
R21
C21
−
+
C12
R12
R22
C22
C3
R3
C4
Figure 2.5: Schematic of the amplifier circuit
Each step acts as a bandpass filter, where C1xand R1xof each step regulates the high-pass cut-off frequency and C2x and R2x regulates the low-pass cut-off frequency. These two cut-off frequencies are both given by [20]
fc= 1
2πRC, (2.4)
where fc denotes the 3 dB-cutoff point. During the measurements an additional filter step was added in order to filter out noise and DC that was still present; this step consists of the components C3, R3 and C4, where C3 blocks DC and R3 and C4 has a low pass cut-off frequency that follows Equation (2.4). The amplification in each step is given by [20]
Gain = −R2x
R1x
.
and the total amplification is given by multiplying the amplification from each step.
2.3.1 Component values
For the plots presented in Chapter 4 – Results from Measurements on the RotaBend Rig, there were two amplifiers used: AMP1 and AMP2. AMP2 is an older version which had a theoretical low pass cut-off frequency at around 1 MHz, a theoretical high pass cut-off frequency at 160 Hz and a theoretical amplification of 100 V/V in each OP-amp step, resulting in a total amplification of 10 000 V/V. The passive filter on the output was not used; this filter was added as an on-site fix to AMP1, as explained in the following paragraph.
Since the test rig created a huge noise peak at around 10 kHz AMP1 was designed. It has a theoretical high pass cut-off frequency of 160 Hz, a theoretical low pass cut-off frequency of 33.9 kHz in each step and a theoretical low pass cut-off frequency from the passive filter of 15.9 kHz. It has the same amplification as AMP2. AMP1 was designed and constructed on the test site with a limited amount of components available. The values were not the ideal but served the purpose, meaning that the 10 kHz noise was suppressed. Table 2.2 shows the component values for AMP1. To be noted is that these values are not theoretically sufficient to reduce the 10 kHz noise, but using real components with error margins in their values created a filter that was sufficient enough to reduce the noise, as shown in Section 4.2.
Table 2.2: Component values for the produced amplification circuit, AMP1
Component Value Unit
C11 1 µF
R11 1 kΩ
C21 47 pF
R21 100 kΩ
C12 1 µF
R12 1 kΩ
C22 47 pF
C3 1 µF
R3 1 kΩ
C4 10 nF
These values were used for spice simulations using the OrCAD Pspice 16.5. Figure 2.6 shows the gain and the two cut-off frequencies for the amplifier circuit in the 0 - 1 MHz frequency range. As shown in the plot, the gain is close to 10 000 (9700) in the midband
2.3. Electronics 15
region, the lower cut-off frequency is a bit higher than the theoretical 160 Hz (240 Hz) and the higher cut-off frequency is a bit lower than the theoretical 15.9 kHz (12.4 kHz). The discrepancy between the simulated cut-off frequencies and their calculated counterparts comes from the simulation using more powerful models to produce the results while the calculations uses the ideal case scenario and simplifications.
Figure 2.6: Amplification and cut-off frequencies for the produced amplifier AMP1
2.3.2 PCB-design
In order to produce a stable and reproducible amplifying circuit a Printed Circuit Board (PCB) was designed. The aim of the design was to make a board of convenient size with connectors for quick and easy connections and disconnections to the sensors and the DAQ board. The choice for input port fell on a female SMA-connector, because it is often used in antenna connections and cables with male/male connectors are easy to find in a great variety. The input to the DAQ board is a female BNC coaxial connector and because male/male BNC coaxial cables are plenty abound it was chosen to have a female BNC coaxial connector as the output from the amplifying circuit. The battery
connections were made with 9 V battery clips. All resistors and ceramic capacitors were of SMD type and size 0805. The PCB design is shown in Figure 2.7. Note that the last filtering step is not included in the PCB, since it was a result of on-site modifications while performing the tests on the rotating bend rig.
Figure 2.7: Amplifying circuit PCB design. Lengths are in mm
2.4 DAQ board
To log and analyse the signal from the amplifying circuit the signal needs to be sampled, using an interface between the amplifying circuit and a computer. The logging and analysis should preferably be possible to perform in the same program or interface.
The signal should as stated in the design goal be of frequencies up to 1 MHz. The Nyquist-Shannon theorem states that in order to fully reconstruct a sampled signal the sample rate has to be twice as high as the highest frequency in that signal. Meaning that in order to reach the specifications, a sample rate of at least 2 MS/s (MegaSamples per second) is needed. A device that fulfills the requirements is a DAQ board (Data Acquistion), which interfaces with a computer to acquire analog data for logging and/or analysis.
The DAQ board used in this work was Measurement Computing’s USB-1602HS, pictured (taken from its user’s guide [22]) in Figure 2.8. This DAQ board meets the requirements of having a sample rate of 2 MS/s and being compatible with Matlab’s DAQ-toolbox, which allowed for logging and analysing in a previously known program. Additionally, it has a high resolution, 16-bit, and software selectable input ranges ± 10 V, ± 2.5 V or
± 500 mV.
2.5. Acquisition function on the PC 17
Figure 2.8: Measurement Computing’s USB-1602HS DAQ
The sampling was done in bursts of a chosen length, and at a pre-set interval. There were two main reasons for choosing a burst based sampling before a continuous one: the amount of data is smaller, and bursts allows for a easier comparison over time by taking the averages of each burst and comparing them. The most commonly used values were a burst length of 1 s (at 2 MHz, giving 2000000 values) and at an interval of 5 s. These values were chosen in order to not miss too much between bursts whilst not producing a unmanageable amount of data.
2.5 Acquisition function on the PC
The interface with the DAQ board to make it perform the acquisition and log the data was made by writing a Matlab function AquireData(SampleRate, duration, interval, range, LogName). When started this function runs until a key is pressed, and outputs a log-file containing data and other information for each acquisition burst.
When the function ends it creates a text document containing the starting time, the sampling rate, the burst length, the interval, the voltage range and the ending time.
The five inputs to the function, SampleRate, duration, interval, range and LogName regulates the following.
SampleRate
The SampleRate-value is the, as the name suggests, rate in Samples/second (S/s)
of the sampling. It has a minimum value of 0.01 S/s and a maximum of 2 MS/s.
duration
The duration-value is the time duration of one sample burst. The minimum value is dependent on the value of SampleRate, in such a way that the product SampleRate*duration can not be less than 1, otherwise the acquired data (if any) will not be from an entire sample period. The maximum value is limited by the available memory on the PC.
range
The range-value is the voltage range in V of the input, selectable as [-500m,500m], [-2.5,2.5] and [-10,10]. It specifies the range over which the resolution is divided.
It should be as closely matched to the measured signal as possible but always be higher than the maximum value of the signal to avoid saturation.
interval
The interval-value sets the time to wait in seconds between bursts, as a non- negative real number between 0 and infinity. This interval is implemented in the function with Matlab’s pause-command executed before the function loops. Due to overhead and computational time when logging burst data the actual interval will be about 1 s longer than the set interval-value.
LogName
The LogName-value is the name that the log files are marked with. A log file is stored for each sample-burst, marked with the date and time that the sampling started along with the LogName string and the burst number ”(0)XX”, in the format: YYYYMMDDTHHMMSSLogName(0)XX. The zero is appended by Matlab when adding the burst number if the LogName’s last character is a letter. If it ends with a number the burst number will be increased from that number. A measurement starting with a LogName ending in a letter will have no burst number for the first measurement and ”01” for the second. For example, the third burst in a sampling started the 6th of June 2014 at 10:25:05 with the name ”TestName” is named 20140605T102505TestName02.
These log-files are saved in the DAQ-toolbox format ’.daq’. In this file is, apart from the logged values (the acquired voltages), both the acquisition object informa- tion and the hardware information stored. Meaning that all values and parameters used during the acquisition (such as sample rate, number of samples, starting time, information about the used hardware, etc.) can be obtained by opening it with the daqread(’filename’,’info’) command.
2.5. Acquisition function on the PC 19 2.5.1 Implementation of AquireData
The underlying Matlab code for the function AcquireData is shown here with the im- portant parts in pure code and the non-critical parts replaced with comments.
The program flow is as follows:
• A routine to stop the program when a key is pressed is initialized; it has to use a graphic window to detect the key press since Matlab has no support for key presses in the main window.
• The text file containing acquisition information and start time is set up.
• The acquisition object (the DAQ board) is initialized with the chosen values.
• The acquisition is performed in an infinite loop, which only ends if a key is pressed.
• If a key is pressed the stop time is added to the text file and the program is ended.
function AcquireData{SampleRate,duration,interval,LogName}
%% Set up keypress to end program %%
global KEY_IS_PRESSED KEY_IS_PRESSED = 0;
gcf;
set(gcf,’KeyPressFcn’,@myKeyPressFcn); %myKeyPressFcn changes the value of KEY_IS_PRESSED from 0 to 1
%% Set up text file log %%
% Load system time
% Create filename from ’LogName’
% Open file and write values to it
%% Set up acquisition parameters %%
ai = analoginput(’mcc’); % Use mcc DAQ board
addchannel(ai,0); % Use channel 0
set(ai,’SampleRate’,SampleRate) % Set sample rate to ’SampleRate’
requiredSamples = floor(SampleRate*duration); % Number of samples required for chosen duration
set(ai,’SamplesPerTrigger’,requiredSamples); % Set ^
waitTime = duration*1.1+0.5; % Wait time for acquisition to be done ai.Channel.InputRange = range; % Set input range
set(ai,’LoggingMode’,’Disk&Memory’);% Log to disk and memory set(ai,’LogToDiskMode’,’Index’); % Index logfile (no overwrite) set(ai,’LogFileName’,LogName); % Set logfile name to LogName
%% Loop for acquisition %%
while ~KEY_IS_PRESSED % Runs until key is pressed
start(ai); % Start acquisition
wait(ai,waitTime); % Wait until done
disp(’Acquiring data...’) % Show something is happening data = getdata(ai); % Load data
%% View fft plot (to show something happening) %%
pause(interval); % Time between bursts end
%% Add timestamp when program ended add to and close text file log %%
disp(’Program ended’) % Show that program ended
delete(ai); % Remove DAQ object
clear ai % -"-
end
C HAPTER 3
Test Setup and Measurement Analysis
This chapter describes the test rig that the fatigue tests were performed on and the signal analysis that was performed on the acquired data.
3.1 Rotating Bend Rig
The tests were performed on a test rig called a ”Rotating Bend Rig” (Power Rotabend by Sincotec). The application of this rig is to evaluate samples of metal. This is done by applying a moment to a metal sample shaped as a round rod, thus bending it and then rotating it until it breaks or until a preset number of revolutions is performed. These rods can either be completely straight or have a thinning waist in the middle; the waisted ones are used when an accelerated fatigue is wanted. A sketch of the two different rods is shown in Figure 3.1. For these tests the rods with the thinning waist were used.
The choice of RotaBend as the test rig instead of measuring on a rolling bearing was made for two main reasons. It allows to measure fatigue in a short time span – the longest measurement running until breakage took three hours, as opposed to running a rolling bearing to breakage which can, even during extremely heavy loading, take days.
Its construction also allows for easy mounting of the sensor close to the point where most stress is prevalent and therefore where most fatigue should arise.
21
(a) Straight rod (b) Rod with waist
Figure 3.1: Sketches of the two rod varieties used by the RotaBend rig
The sensor was mounted inside the rig with the rest of the measuring system in an adjacent room in order to reduce the noise that was present near the rig. An image of the sensor mounted in the test rig shown in Figure 3.2. In this image the sensor is mounted in what is called the perpendicular orientation, called so since it is perpendicular to the length of the rod. Measurements were also performed with the sensor rotated 90 degrees, so that it is parallel to the length of the rod. This orientation is therefore called the parallel orientation.
When a rod is bent in the fashion it is in the RotaBend rig, into a bow like shape, the maximum tensile stress appears in a point at the middle of the outer curve, the side facing towards the reader in Figure 3.2a, and the maximum compressive stress appears in the corresponding point in the inner curve. Measurements with the sensor placed in either of these points might pick up more BN, but are not possible to perform since the rod swings rapidly back and forth when breakage occurs and would then break the sensor. Since the rod is rotating, the points of maximum stresses are not static, so the fatigue can still be measured.
(a) Frontal view (b) Top view
Figure 3.2: Sensor mounted in the RotaBend test rig
The rods used in the tests were leftovers from other tests and there was no possibility to use rods of the same material for all tests. Therefore a full comparison of all sensors and directions could not be performed. There were however tests performed on the same
3.2. Signal processing and information extraction 23
specimen with different sensors for a few rotations, in order to provide some comparison data.
3.2 Signal processing and information extraction
The data from the measurements on the specimens mounted in the RotaBend rig was analysed using two different methods: calculating the Fourier transform for each burst, and calculating the RMS voltages for the whole measurement. The Fourier transform method was chosen to see if there was any change in the frequency content as the rods were fatigued, and the RMS voltages method to see if there was any change in the signal strength as the rods were fatigued.
3.2.1 Fourier transform
A Fourier transform is a transform that expresses a function of time as a function of frequency. This transformation makes it possible to see the frequency content of a sampled signal. Take for example the function f3(t) = f1(t) + f2(t) = sin(2πt) + 0.2 sin(10πt) shown in Figure 3.3a. As can be seen, the function consists of two sinusoids of amplitude 1 and 0.2, which are also shown in the figure. Looking only at f3, it is hard to tell what the frequency contents are, but when taking the Fourier transform of the function, shown in Figure 3.3b, it is clear that the function has two frequency components, one with amplitude 1 at 1 Hz and one with amplitude 0.2 at 5 Hz.
0 1 2 3
−1
−0.5 0 0.5 1
t (s)
y=f(t)
f1(t) f2(t) f3(t)
(a) Wave function f3(t) and its two compo- nents
0 1 2 3 4 5
0 0.5 1
Frequency (Hz)
Amplitude
(b) Fourier transform of f3(t)
Figure 3.3: Example of a Fourier transform of a wave function
The use of the Fourier transform on the measured data allows the capability to see if there are any frequency changes over the lifespan of a specimen. It also allows for a comparison of the frequency content between different specimens, and a comparison between measurements with different sensors. Since the logged data is not a continuous signal, but a finite number of values, the Fourier transform cannot be used directly.
Instead a Discrete Fourier Transform (DFT) is used, which works in the same way as the Fourier transform, but for discrete values.
3.2.2 RMS voltage
For one burst of n sampled values, the RMS value is the square root of the arithmetic mean of the squares of the original values. So for a set of voltages {V1, V2, . . . , Vn} the RMS voltage is
VRMS = r1
n(V12+ V22+ · · · + Vn2). (3.1) The RMS voltages were calculated for each burst and plotted against the burst number, which gives a graph in which any change in the total received signal can be seen. An ex- ample of how such a plot could look is shown in Figure 3.4, which shows the hypothetical RMS voltages for 6 consecutive bursts.
3.2. Signal processing and information extraction 25
1 2 3 4 5 6
Burst number VRMS
Figure 3.4: Example of RMS voltage plot for 6 bursts
Apart from the RMS voltages from the raw measurement data, there were four other RMS voltage plots produced. These were RMS voltages calculated on frequency filtered data, in four different spans: 0-1.8 kHz, 1.8-3.75 kHz, 3.75-4.5 kHz, 4.5-1000 kHz. The spans were chosen by looking at the Fourier transform plots and identifying areas where there might be activity that differs from the other areas. The span divisions are shown in Figure 3.5. As can be seen, there are some peaks in each division, and the reason to isolate them is to see if there is any change in the signal in that specific frequency span and to see if this change can be used to identify fatigue. This could also be helpful in order to determine in what frequency ranges the BN is most prevalent.
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
1.8 kHz 3.75 kHz
4.5 kHz
Frequency
Amplitude
Figure 3.5: Frequency span division shown on the Fourier transform plot from a measurement on a 7P2 specimen using Sensor 1 parallel to it
The filtering was performed by taking magnitude values in a frequency span from the magnitude value vector created by the Fourier transform. The RMS values was then calculated from that data. This is not a true filtering, but was done in order to reduce computational time. A true filtering using a bandpass Butterworth filter takes up to 20 minutes to perform a filtering of one span. To show that the results would look the same by using this ’filtering’ method and by using a real filtering method the plots in Figure 3.6 were produced. As can be seen the shape of the plots is the same, in which we are interested, but the amplitudes are different. The amplitude difference comes from the way the Fourier transform is calculated, and can be fixed by normalising the results from the transform, this was however deemed unnecessary since only the shape of the voltages over time is of interest in this work, and not the amplitudes.
0 100 200 300
1.85 1.9 1.95
2 ·10−2
(13, 0.01997)
(281, 0.01822)
Burst number VRMS
(a) Filtering using Fourier transform data
0 100 200 300
0.52 0.54
0.56 (13, 0.5666)
(281, 0.5107)
Burst number VRMS
(b) Filtering using a Butterworth bandpass fil- ter
Figure 3.6: Comparison between different filtering methods
3.2.3 Implementation in Matlab
The RMS voltage and Fourier transforms were calculated from the logged data using Matlab. The Matlab code is presented with some parts shown only as comments. The fftdaq-function uses the text string to load the voltages from the LogName.daq file, which are the voltages from the first burst. It then uses the voltages to calculate the DFT, using an algorithm called the Fast Fourier Transform (FFT). The frequency span for the FFT, the FFT absolute values and the original voltages are then returned to the
3.2. Signal processing and information extraction 27
program, where the FFT is plotted.
The original voltages are used to calculate the RMS voltage for the first burst, using Equation (3.1). The RMS voltages for the subspans are calculated by taking the values from the FFT-amplitude data vector. Note that the vector positions do not equate to the frequencies, so the positions had to be cross-referenced from the frequency vector f0.
It was done by hand for this work, but it can also be done (with increased computational time) by using Matlab’s find function. For the rest of the burst the calculations are performed in a loop going through all samples.
[f0,Y0,data] = fftdaq(’LogData’,2000000); % Load first burst data
% Plot FFT
dataRms = zeros(samples,1); % Initiate Rms-voltage vector
% V_1, V_2, V_3, V_4 subspans also initiated
dataRms(1) = sqrt(sum(data.^2)/length(data)); % Calculate RMS-voltage for first burst, put in RMS-voltage vector
% Subspans first RMS-voltage also calculated and put in each vector :%
V_1(1,1) = sqrt(sum(Y0(1:1889).^2)/length(Y0(1:1889))); % 0-1800Hz span V_2(1,1) = sqrt(sum(Y0(1889:3934).^2)/length(Y0(1889:3934))); % 1800-3750Hz
span
V_3(1,1) = sqrt(sum(Y0(3934:4720).^2)/length(Y0(3934:4720))); % 3750-4500 Hz span
V_4(1,1) = sqrt(sum(Y0(4720:end).^2)/length(Y0(4720:end))); % 3750Hz - end span
for i = 1:samples-1 % Loop through all bursts
filename = sprintf(’LogData0%d.daq’,i); % File to load [~,Y,data] = fftdaq(filename,2000000); % Load burst data
dataRms(i+1) = sqrt(sum(data.^2)/length(data)); % Calculate RMS-voltage
% Calculate subspans RMS-voltage end
% Plot RMS-voltages
%--%
function [f,absY,data] = fftdaq(file,SampleRate) data = daqread(file); % Load sampled voltages L = length(data(:,1)); % Length of vector NFFT = 2^nextpow2(L); % Number of FFT-points Y = fft(data(:,1)’,NFFT)/L; % Calculate the FFT
absY = 2*abs(Y(1:NFFT/2+1)); % Present as absolute values
f = SampleRate/2*linspace(0,1,NFFT/2+1); % Frequency vector for the FFT values end
C HAPTER 4
Results from Measurements on the RotaBend Rig
In this chapter the performed measurements are described more thoroughly and the most important and interesting plots are presented. A full compilation of all plots is provided in Appendix A, to the benefit of the interested reader.
4.1 Performed measurements
There was a total of 16 life span tests performed, each on a different specimen (rod).
A life span test is the denomination used in this work for a test where the specimen is rotated until it breaks, and during which the applied moment and the rotational speed stays constant. All tests, unless otherwise stated, were performed with a rotational speed of 3000 rev/min. On two of the specimens, one of 7P material and one of 8R3 material, speed and sensor tests were performed prior to the life span tests. These tests were performed with a small load (15 or 30 Nm) and should not have affected the life span results. Table 4.1 shows a compilation of the specimen materials, the number of specimens, the sensor used when measuring on those specimens and the orientation of the sensor. Further explanations of the tests are presented in the following sections.
29
Table 4.1: Specimen material, number of each specimen material, sensors and their orientation used upon specimens for the measurements. ’p’ is parallel orientation, ’pp’ is perpendicular orientation and ’-’ is sensor not used. Bold is orientation used for life span measurement.
Material Number of Sensor
specimens 1 2 3 4
8R3 4 p,pp - - -
7P2 3 p,pp pp - -
7P 3 p,pp p,pp p,pp -
7AXII 2 - - pp -
6AX 1 - - pp -
3MS1 1 pp - - -
Grad3MSI 2 pp - p,pp p,pp
4.2 Difference between AMP1 and AMP2
In order to compare the two amplifying circuits AMP1 and AMP2 there were four lifespan measurements performed with Sensor 1 on the specimens of 8R3-material. Two of these measurements were performed using AMP2 as the amplifier, which let through a lot of noise from the test rig, and the other two used AMP1, which was redesigned to block the noise from the test rig. The applied moment for the AMP2 measurements was 40 Nm. This gave a very long test time before the specimens broke, and was therefore changed to 50 Nm for the AMP1 measurements. The burst length for the measurements with AMP2 was 1 s and the interval was 20 s. The interval was decreased to 1 s for the AMP1 measurements, with the burst length the same, because the specimens were expected to break faster with a higher load.
4.2.1 RMS voltages
The full scale RMS voltages from the measurements are shown in Figures 4.1. As can be seen it is easier to see a change during the lifespan for the measurements with AMP1.
For the measurements with a parallel sensor it might be easier to see a change when removing the peak in the beginning of the measurement. This will be investigated in Section 4.6.
4.2. Difference between AMP1 and AMP2 31
0 50 100 150 200 250
0.76 0.78 0.8 0.82 0.84
Burst number VRMS
(a) AMP1 (pp). Total runtime: 0:9:57
0 50 100 150 200 250
0.47 0.48 0.49 0.5
Burst number VRMS
(b) AMP2 (pp). Total runtime: 2:58:10
0 100 200 300 400
0.79 0.8 0.81 0.82 0.83 0.84
Burst number VRMS
(c) AMP1 (p). Total runtime: 0:16:44
0 50 100 150 200 250
0.45 0.5 0.55
Burst number VRMS
(d) AMP2 (p). Total runtime: 1:47:57
Figure 4.1: Comparison of RMS voltages amplified with AMP2 and AMP1. Measurements per- formed with Sensor 1 on a 8R3 specimen. p is parallel orientation, pp is perpendicular orientation
4.2.2 Fourier transforms
The differences in the Fourier transforms for measurements with AMP1 and AMP2 are shown in Figure 4.2, where it can be seen that the 10kHz peak detected with AMP2 is almost entirely gone when using AMP1. It can also be seen that the rest of the values are much higher, which is a result of the Fourier transform presenting the amount of each frequency that is present in a signal and when a part is removed the remaining are shown as larger.
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(a) AMP2
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(b) AMP1
Figure 4.2: Comparison between AMP1 and AMP2, using Sensor 1 perpendicular to a 8R3 specimen
4.3 Identifying a difference over time using the Fourier trans- form
Looking at the Fourier transforms of the bursts give little visual information about the fatigue during the lifespan. Take for example the plots in Figure 4.3, which show the Fourier transform of three different measurement bursts from a measurement on a 8R3 specimen. The bursts being the first, when the measurement just started; the middle, halfway through the measurement; and the last before the specimen broke. The frequency content in these plots are very similar, and the small difference that can be seen lies in the amplitude, a difference that is easier to compare burst-for-burst by using the RMS voltages for each burst.
4.3. Identifying a difference over time using the Fourier transform 33
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(a) First burst
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(b) Middle (103rd) burst
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(c) Last (206th) burst
Figure 4.3: Comparison between Fourier transforms of three measurement bursts from three points in the lifespan: the first burst, the middle of the lifespan burst and the last before stopping burst. The bursts were performed on a 8R3 specimen with Sensor 1 perpendicular to it
To further add to this argument, plots from a measurement on a 7P2 specimen were produced, shown in Figure 4.4. The same observation as earlier can be made for these bursts – they show no easily detectable difference in the frequency content, but a small difference in amplitude.
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(a) First burst
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(b) Middle (130th) burst
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(c) Last (260th) burst
Figure 4.4: Comparison between Fourier transforms of three measurement bursts from three points in the lifespan: the first burst, the middle of the lifespan burst and the last before stopping burst. The bursts were performed on a 7P2 specimen with Sensor 1 perpendicular to it
When comparing the first burst of each measurement on the 8R3 and 7P2 specimens there is no easily detectable frequency difference either; the major difference that can be seen lies in the amplitude. This is also shown when comparing to the plots in Figure 4.5
which shows the Fourier transforms of the first burst from measurements on a 6AX and a 3MS1 specimen.
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(a) 6AX
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(b) 3MS1
Figure 4.5: Comparison of Fourier transforms of the first burst from measurements on a 6AX and a 3MS1 specimen measured with Sensor 1 perpendicular to them
4.4 Frequency content when measuring away from the spec- imen or with stopped test rig
In order to show the frequency content of the background noise and the noise that the RotaBend rig created when running, there were two tests performed with a sensor placed in the rig but away from the specimen. The first measurement was performed with the rig stopped, and the other with a 8R3 specimen loaded with 30 Nm and the rig running at 3000 rev/min. Fourier transforms was created from both these tests, shown in Figure 4.6. As can be seen there is almost no noise present when the rig is stopped, apart from a peak at about 6 kHz. The cause of this peak can be external noise; looking at the other presented Fourier transforms (Figure 4.3, 4.4 and 4.5) it seems like it does not appear when measuring on the specimens and is therefore not interfering with the measurements. Looking at the plot from the measurement performed when the rig was running there is more noise present, a lot especially in the 4 kHz area. This noise is also seen in the other Fourier transform plots (Figure 4.3, 4.4 and 4.5) and is most likely caused by noise from the rig.
4.5. Comparison between full span RMS voltages for different specimens 35
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(a) Test rig stopped
0 0.5 1 1.5
·104 0
0.1 0.2 0.3 0.4
Frequency
Amplitude
(b) Test rig running at 3000 rev/min with 30 Nm load
Figure 4.6: Fourier transforms of measurements performed with Sensor 1 placed in the test rig but away from the specimen
4.5 Comparison between full span RMS voltages for differ- ent specimens
This section presents the full span RMS voltages from the life span measurements. The plots for these comparisons are sorted in sections by the material of the specimens. The 8R3 plots have already been shown in Figure 4.1. A discussion of these plots is carried out in Chapter 5 – Conclusions and Discussions.
4.5.1 7P2
There were three lifespan measurements performed on the specimens of 7P2-material.
All were amplified with AMP1 and loaded with 50 Nm. Two of the measurements used Sensor 1, in perpendicular and parallel orientation, and one used Sensor 2 in per- pendicular orientation. The burst length for the measurements with Sensor 1 was 1 s and the interval was 1 s. This interval was found to be unnecessarily short, so for the measurement with Sensor 2 it was increased to 5 s. The burst length was kept at 1 s.
Figure 4.7 shows the full scale RMS voltages from the measurements on the three speci-
mens. Comparing these, it can be seen that the parallel Sensor 1 and the perpendicular Sensor 2 measurements create similar looking plots, where the values show a tendency to quickly increase and then decrease, whereas the perpendicular Sensor 1 measurement only increases and then levels out.
0 100 200 300
0.74 0.76 0.78
Burst number VRMS
(a) Parallel Sensor 1. Total runtime:
0:14:11
0 100 200 300
0.92 0.94 0.96 0.98 1
Burst number VRMS
(b) Perpendicular Sensor 1. Total runtime:
0:13:09
0 50 100 150
0.8 0.82 0.84 0.86
Burst number VRMS
(c) Perpendicular Sensor 2. Total runtime:
0:20:10
Figure 4.7: Comparison of full scale RMS voltages from measurements with Sensor 1, perpen- dicular and parallel, and Sensor 2 perpendicular to three different 7P2 specimens
4.5.2 7P
There were three lifespan measurements performed on the specimens of 7P material.
All were amplified with AMP1 and loaded with 50 Nm. Two of the measurements
4.5. Comparison between full span RMS voltages for different specimens 37
used Sensor 1, both in the perpendicular orientation, and one used Sensor 3 in parallel orientation. The burst length was 1 s and the interval was 5 s for all measurements. The two measurements performed with Sensor 1 were made to see if there is any repeatability between tests on same material specimens.
Figure 4.8 shows the full scale RMS voltages from the measurements on the three speci- mens. There are similarities in all the plots, where the values show a slight tendency to increase, decrease and then increase before the specimens break.
0 20 40 60 80
0.94 0.96 0.98 1
Burst number VRMS
(a) Perpendicular Sensor 1, one. Total run- time: 0:9:14
0 20 40 60 80 100
0.9 0.92 0.94 0.96
Burst number VRMS
(b) Perpendicular Sensor 1, two. Total run- time: 0:11:10
0 50 100 150
0.52 0.54
Burst number VRMS
(c) Parallel Sensor 3. Total runtime:
0:16:39
Figure 4.8: Comparison of full scale RMS voltages from two perpendicular Sensor 1 measure- ments and one parallel Sensor 3 measurement on three different 7P specimens
4.5.3 7AXII
There were two lifespan measurements performed on the specimens of 7AXII material.
Both were amplified with AMP1 and both of the measurements used Sensor 3 perpen- dicular to the specimen. The difference between the two was the applied moment, which was 40 Nm versus 50 Nm; this moment change was made due to the fast breakage of the specimen when loaded with 50 Nm. Both measurements had a burst length of 1 s and an interval of 5 s.
Figure 4.9 shows the full scale RMS voltages from the measurements. There is a clear change seen for the 40 Nm case, the values increase quickly and then slowly decrease.
For the 50 Nm the specimen broke too quickly to get any useful data. What can be said when comparing the two is that a higher applied moment seems to give higher RMS voltages.
0 100 200 300
0.8 0.85
Burst number VRMS
(a) 40 Nm applied moment. Total runtime:
0:29:57
0 10 20 30
0.95 1 1.05 1.1
Burst number VRMS
(b) 50 Nm applied moment. Total runtime:
0:4:42
Figure 4.9: Comparison between applied moments on 7AXII specimens, measured with Sensor 3 perpendicular
4.5.4 6AX
There was one lifespan measurement performed on a 6AX specimen. It was amplified with AMP1 and used Sensor 1 perpendicular to the specimen. The burst length was 5 s and the interval was 5 s for this measurement. The applied moment was 45 Nm, chosen halfway between the two previously used with the hope of getting a good measurement
4.5. Comparison between full span RMS voltages for different specimens 39
out of the only specimen.
Figure 4.10 shows the full scale RMS voltages from the measurement. As can be seen there is a large peak in the beginning that makes change over time appear small. From what can be seen the RMS voltages plot seems to behave like the other materials RMS voltages plots; it increase quickly in the beginning and then level out.
0 50 100 150 200 250
0.80 1.00 1.20 1.40
Burst number VRMS
Figure 4.10: Full scale RMS voltages from a perpendicular measurement on a 6AX specimen with Sensor 1. Total runtime: 0:26:16
4.5.5 3MS1
There was one lifespan measurement performed on the 3MS1 specimen. It was amplified with AMP1 and used Sensor 1 perpendicular to the specimen. The burst length was 1 s and the interval was 5 s for this measurement. The applied moment was 45 Nm, chosen for the same reason as for the 6AX specimen, to hopefully get a good measurement out of the only specimen.
Figure 4.11 shows the full scale RMS voltages. As can be seen, the voltages increase quickly and then level out. For the final bursts there seems to be a slight increase before the breakage of the specimen.
0 20 40 60 80 100 120 0.82
0.84 0.86 0.88 0.9
Burst number VRMS
Figure 4.11: Full scale RMS voltages from a measurement perpendicular on a 3MS1 specimen with Sensor 1. Total runtime: 0:14:20
4.5.6 Grad3MSI
There were two lifespan measurements performed on the Grad3MSI specimens. They were amplified with AMP1 and used Sensor 1 and Sensor 4 perpendicular to the spec- imens. The burst length was 1 s, the interval 5 s and the applied moment 45 Nm for both measurements.
Figure 4.12 shows the full scale RMS voltages for the measurements. They share some similarities in the beginning of each where they increase, but most notable is the increase in amplitude shown for Sensor 4, which gives RMS voltages almost twice as high as Sensor 1.
