of an industrial line
Ignacio Gutierrez
Maintenance Engineering, master's level (120 credits) 2020
Luleå University of Technology
Department of Civil, Environmental and Natural Resources Engineering
1
Table of content
1 Abstract ... 3
2 Introduction ... 4
2.1 Aim ... 4
2.2 Objectives... 5
2.3 Limitations... 5
3 Theoretical framework: Prognostics & Health Management (PHM) ... 6
3.1 Industrial maintenance types ... 7
3.2 Condition based monitoring techniques... 7
3.3 Health prognostics ... 10
3.4 System taxonomy ... 12
3.5 Failure mode effects and criticality analysis – FMECA ... 12
3.6 Fault Tree Analysis – FTA ... 15
3.7 Reliability Block Diagram ... 16
3.8 Rotating machine diagnosis ... 17
3.8.1 Gear Mesh Frequency ... 17
3.8.2 Bearing failure frequencies ... 18
4 Method ... 20
5 Results and discussion ... 21
5.1 Technical Breakdown: System enumeration ... 21
5.2 FMECA: Potential failure modes identification ... 21
5.3 Frequency analysis ... 24
5.3.1 Gear mesh frequency ... 24
5.3.2 Bearing frequencies ... 24
5.4 Acquisition system: Identification and monitoring of failures ... 24
5.4.1 System topology ... 25
5.4.2 Sensor selection ... 27
5.4.3 Implementation architecture ... 29
5.4.4 Connexion ... 32
5.5 Data analysis plan ... 33
5.5.1 Strategy ... 33
5.5.2 Pre-processing techniques ... 33
5.5.3 Graphical results ... 33
6 Conclusion ... 36
2
7 References ... 37
8 Appendix ... 39
8.1 FMECA ... 39
8.2 FTA ... 42
3
1 Abstract
Industry 4.0 push forward the development of concepts such as artificial intelligence, big data, and Industrial Internet of Things, which claims an evolution of the monitoring systems design in terms of the accessibility to the information. In this project, the author describes the design of a condition monitoring system to monitor the state of different components of an extrusion line and propose a system that allows predictive maintenance in industry, specifically an extrusion system.
In this framework, Condition Based Maintenance, CBM, the health state of the component is continuously monitored. In some cases, the monitoring can be periodical. The goal is to make repairs in the opportune moment, by receiving data of malfunction, so the efficiency is maxed.
The aim is to develop a system to monitor the state of different components of an extrusion line and to propose a system that allows predictive maintenance in industry and that the project can be used as a guideline to a complete condition monitoring system implementation in an industrial environment.
To be able to achieve this, some first steps must be accomplished. These are a Taxonomy, or a breakdown of the system into individual elements and how they are related; and carrying out a Failure Modes Effect and Criticality Analysis, known as FMECA. With these studies, the author displays the failure modes that are critical for the operating of the system and thus, which have the most likelihood to occur while having a big impact.
From the information extracted the author presents a model based in accelerometer, temperature and MCSA (Motor Current Spectral Analysis) sensors. Furthermore, the data obtained will need to be analysed. With that in mind, the operating frequencies as well as the failure modes frequencies must be studied, which will allow correct identification while analysing the data. This analysis will be done by characterizing data and applying analysing techniques as FFT or Hilbert.
4
2 Introduction
With Industry 4.0 completely stablished in our industrial tissue, many companies require up to date solutions to keep being competitive. With the growing needs of having excellence in production and taking advance of the improvements in the communication technologies, the maintenance techniques have been evolving. Maintenance is “the combination of all technical and administrative actions, including supervision actions, intended to retain an item in, or restore it to, a state in which it can perform a required function.” [1]
Reliability and maintainability engineering attempts to study, characterize, measure, and analyse the failure and repair of systems. In order to improve upon their operational use by increasing their design life, eliminating or reducing the likelihood of failures and safety risks, and reducing downtime thereby increasing available operating time.
This project is based in an extrusion system in the food industry, specifically in the extruder head and supply conveyor belts. From this system, an approach to an industrial CBM system implementation will be studied. CBM uses condition monitoring techniques to determine whether a problem exists in equipment, how serious the problem is, and how long the equipment can run before failure. The basic principle of this technique is to use online sensing tools to obtain the current product degradation data and minimize the system downtime by balancing the risk of failure and achievable profits [2].
To understand how the system operates, it works non-stop except for caliber changes. This means that, if there is no tool change planned, the machine works 24 hours 7 days. Obviously, this is a highly demanding process, and losing time for an unplanned repair can be fatal. After the extrusion process is completed, a conveyor belt collects the product and transports it through the different areas until the end of the line.
The components susceptible to predictive maintenance are the following:
• Extruder heads:
o Bearings o Pinions
• Extruder engine:
o Imbalance o Rotor faults o Air gaps
• Pull engines of the conveyor belts
o Imbalance regarding the torque exercised o State of health of the engines
2.1 Aim
The aim in this proposal is to develop a system to monitor the state of different components of an extrusion line and to propose a system that allows predictive maintenance in industry.
Furthermore, the purpose is to be used as a guideline to a complete condition monitoring system implementation in an industrial environment, from system weakness identification to data analysis, going through decision taking. After the implementation of the system, the industrial company will have a notorious improvement of availability to consider the project successful.
5 To do so, a study of the possible maintenance methodologies must be accomplished, as well as a in depth system study, to be able to advise the best solution.
2.2 Objectives
Following the project aim, walking through the study and implementation of a condition monitoring system, its objectives are:
1. Enumerate the system elements.
2. Identify potential failure modes.
3. Identify the main failure frequencies in order to learn the system symptoms.
4. Propose an efficient way to identify and monitor mentioned failures. This covers both the review of the available sensors and the monitoring strategies as well as topologies to be used.
and the manufacturing of the proposed solution.
5. Develop a data analysis plan.
2.3 Limitations
The main limitation is that is a research project depending on a real manufacturing plant, which implies on the limited information that can be accessed and also timed visits need to be arranged. Related to the actual industry, the only limitation is regarding food industry regulations, that don’t allow the use of some specific materials such as glues, etc. [3]
6
3 Theoretical framework: Prognostics & Health Management (PHM)
Figure 1 PHM progress
Prognostics and health management (PHM) systems aim to increase system reliability, availability, safety and to reduce the maintenance cost of engineering assets[4]. It is based in the following concepts:
• Detection: detecting that there is an anomaly in the behaviour of the monitored component.
• Diagnostics: Diagnosing the type of problem occurring, identifying the affected components.
• Prognostics: Predicting the probability of the monitored component failing within a time frame or estimating RUL.
Applying PHM techniques makes possible to identify when a failure is about to occur, which means no unexpected failures take place. Also allows identification of specific target actions for affected components, which is translated to less troubleshooting time, fewer maintenance interventions, minimal inspections on field and reduced downtime.
Figure 2 Maintenance evolution[5]
7
3.1 Industrial maintenance types
According to [6], importance of maintenance is such that represents an average 4% of the fixed assets costs in various industries. In order to select they type of maintenance to implement several fields must be taken into account. These are, investment required, environmental and safety surrounding, failure costs, estimated failure times, etc. There are different maintenance strategies available:
Figure 3 Maintenance strategies [6]
• Corrective maintenance (CM): it is based on the repair of a component after the fault occurs.
It is used where failure does not have a great impact on successful operation.
• Preventive maintenance (PM): this maintenance strategy is based in fixed time intervals in which components are revised and/or repaired, regardless of the health state, before it reaches the end of its useful life. This strategy is mainly used on components under direct contact to failure factors (corrosion, oxidation, etc.)
• Condition based maintenance (CBM): in this strategy, the health state of the component is continuously monitored. In some cases, the monitoring can be periodical. The aim is to make repairs in the opportune moment, by receiving data of malfunction, so the efficiency is maxed.
3.2 Condition based monitoring techniques
Figure 4 CBM techniques representation
Reliability prediction, traditionally, is based on failure times data. Usually, preventive maintenance strategies don’t consider the current health state of the equipment and base the replacement in time
8 standards. In the other hand, a maintenance strategy developed taking into account the current degradation and evolution of the equipment is referred as Condition Based Maintenance (CBM). The basic principle of this technique is to use online sensing techniques to obtain the current system degradation data and minimize downtime by balancing the risk of failure and profits [2].
Techniques for condition monitoring are mainly related with vibrations and acoustic techniques, focused on failures detection. But the real non-intrusive aim is to develop an integrated system for health assessment of manufacturing machines based on the smart analysis of machine operational conditions. This can be achieved by focusing on current signals of the drives, power consumption of the motors plus available information of the process through the controller, all of them easily accessible. [7]
Current Transformers serve to measure AC amperage in an electrical circuit. They are accurate, safe, easy to implement and reliable[8]. The different condition monitoring methods like vibration or acoustic monitoring usually require expensive sensors. Recent research [9] have surrounded electrical monitoring of the motor, specially focusing on the stator current. Not only to evaluate the motor efficiency concerning energy saving, but also to address problems and failures that could lead to a plant stoppage. The common faults of the induction motors include the rotor eccentricities and misalignment, worn bearings, stator winding insulation faults, broken rotor bars, winding overheating, and load torque oscillations. [9]
Moreover, the utilization of the machine current to avoid tool failures and wear for different production systems is widely studied [10]. The main problem of this type of tool condition monitoring is the consistency of the process, due to the variability. Material changes and short production series make difficult to assess the tool condition since tool wear is not constant with each process and it is difficult to have all processes defined.
MCSA uses the electric motor as a transducer, letting the user to assess the electrical and mechanical condition of the electric motor and which can be extended to the machine. The current signal is modulated due to vibration, as it causes pulses in torque and appears in harmonics. Thus, current spectrum analysis can be used to detect failures in the driven loads. Data processing and analysing mechanisms for early detection of faults in a production machine using current and voltage data[11].
The method used is the Electric Signature Analysis. The basis behind it is that any load and speed variation within an electro-mechanical system produces correlated variations in current and voltage consumption of that system. As a consequence, time and frequency signals reflect loads, stresses, and wear throughout the system, but this requires a pattern recognition. The fault identification is done by comparison between a reference, which will be an electric signature of an equipment in good conditions (the healthy fingerprint), and the equipment under monitoring[11].
Signature Analysis is only applicable to cases in which the principle cause-effect is verified and modelized. So, the first step will be to make a proper design of experiments to characterize the rollers problems to be predicted using the power and current signals.
A high number of measurements must be performed (semisupervised learning) to correlate parameters in order to compare features of abnormal signals of failure with healthy signals. Moreover, an evolutionary trend of the degradation will be determined, this way Signature Analysis can be implemented also to support some prognostic[11].
9
Figure 5 Power based health assessment [12]
In order to obtain the health fingerprint [12], the standard procedure is to run a test cycle, with no load condition, that will generate a better model, and by extend, will have better failure detection.
Working without load will allow to avoid normal machine process noise that could introduced by the load. By employing this method, the different outcomes can be adapted to other similar machines, and it could avoid adaptations problems.
Each machine will have their own attribute saved in their fingerprint. So, the first step will make a first approach of healthy status to train normal condition. After, and this will be done when and where the machine is in operation, the distance between the selected features of the real-time condition monitoring with the healthy status needs to be evaluated. This comparison is made with pattern recognition and unsupervised learning methods.
In summary, the power-based health assessment will consist in:
• Models coming out from the experimental research
• Procedure to have the machine healthy fingerprints
• Testing procedure to be used when the machine is in operation and the pattern recognition algorithms to predict the failure and make the prognosis about when the failure would take place.
Although application experience is presently lacking in areas outside those already cited, it is likely that MCSA will provide a highly sensitive, selective, and cost-effective means for on-line monitoring of the condition of a wide variety of heavy industrial machinery [13]. For example:
• Motor-driven compressors and pumps.
• Rolling mill stands.
• Mixers and crushers.
• Fans and blowers.
• Material conveyors.
Extensive test data support the conclusion that MCSA is a useful tool for monitoring the mechanical and electrical condition of MOVs, particularly in relation to their operational readiness. Experience with motor-driven machinery other than MOVs, though limited, strongly suggest that MCSA is equally applicable to monitoring present condition and to diagnosing impending trouble in a wide variety of consumer and industrial equipment.
10
3.3 Health prognostics
The prognosis plans the health state of equipment into the future. It is able to estimate the remaining useful life (RUL) based on linear and exponential regression of measured values and the according alarm limit.
A prognostic program generally consists of four technical processes or sections: data acquisition, health indicator (HI) construction, health stage (HS) division, and RUL prediction.
Data Acquisition
HI Construction
HS Division
RUL Estimation
Figure 6 Sections examples in a machinery health prognostic program [14]
Measured data, such as vibration or temperature signals, are acquired from sensors, which is known as Data acquisition.
11 HI construction plays an important role in prognostics. From the measured data, signals are processed by signal processing techniques, AI techniques, etc., to represent the health condition of the machinery. An adequate HI is expected to simplify prognostic modelling and produce accurate prediction results. There are two important problems related to the construction of HI. How to build HI from monitoring signals and how to evaluate the suitability of the constructed HI for the RUL prediction. Metrics for the evaluation of prognostic HIs can be classified into five categories:
• Metrics depending on multiple HIs: Monotonicity, robustness.
• Metrics depending on a HI and time: Trendability[14].
• Metrics depending on a HI and HS sequence: Identificability.
• Metric depending on multiple HIs: Consistency.
• Hybrid metrics.
HS division is similar to fault detection or fault diagnosis. But their aims are different. Fault diagnosis is to identify the fault pattern and severity of machinery at a single time point, while HS aims to divide the continuous degradation processes of machinery into HS according to the varying trends of HIs.
Figure 7 Degradation process [14]
The RUL of machinery is defined as the length from the current time to the end of the useful life. “A mechanical system must be composed of multiple components.” The defect of one component may spread to others because of frequent connection between different components. Therefore, it is a significant task to analyse the fault interaction among different components for the RUL prediction of machinery at a system level. [15][16]
12
3.4 System taxonomy
A taxonomy is a procedure that describes how different concepts are related and organized within a specific hierarchical structure.
The based model system is an extruder head for a collagen line. The main stop causes are:
• Heat.
• Mix stuck.
• Caliber changes
Currently, SCADA PLC interface supplies current values for analysis.
The taxonomy is done following the standard ISO 14224.
3.5 Failure mode effects and criticality analysis – FMECA
Figure 8 FMEA breakdown
FMECA is a technique that plays a fundamental role in reliability assurance programs. It is an inductive method that allows a qualitative analysis of the reliability or safety of a system proceeding from bottom to top.
Its purpose is:
• Evaluate the effects and sequences of the events caused by each known failure mode, whatever their cause, to the different levels of the functional hierarchy of the system.
• Determine the importance or criticality of each failure mode, taking into account its influence on the correct operation of the system and its impact on the reliability or safety of the related processes.
• Classify the failure modes identified according to the ease with which they can be detected, diagnose, check, change a component and according to the necessary means to be able to carry out the maintenance of the system.
• Estimate the scales of importance and probability of failure, subject to the availability of the necessary information.
13 Is a method essentially adapted to the study of the failures of materials and equipment, and that can be applied to systems of different technologies or that combine different technologies (electrical, mechanical, hydraulic, etc.). Can be used alone or to complement and support other methods of reliability analysis. It should be applied already from the initial phases of the design of a system or subsystem. It is appropriate for all phases of the design of the system, and the personnel who perform it must be specially trained.
It is based in a system breakdown, or taxonomy, into system structure and start-up, operation, control and maintenance of the system:
System structure includes:
• System elements with its characteristics, behaviour and functions
• Element interconnections
• Level and nature of redundancies
• Localization of the system in the whole installation
Start-up, operation, control and maintenance of the system takes into account:
• Task duration
• Time between periodic tests
• Available time for corrective actions before serious consequences arise
• The complete installation, environment or personnel
• Repair conditions (time, equipment, personnel, etc.)
• Start-up procedure
• Operating control
• Preventive or corrective maintenance
FMECA requires certain modelling of the system, a simplification of the relevant information. Is usual to use symbolic representations of the structure and operation. Normally block diagrams are used to reveal all the essential functions of the system and they contribute to facilitate it understanding.
One of the key concepts of FMECA is the failure mode concept. A failure mode is the effect by which a fault is observed in a system element. It is important to match all the potential failure mode, taking into account the following considerations:
• If the component is new, if it can be referred to similar ones. If the component has already been used, the previous information can be checked.
• Complex components will be breakdown to simpler elements, as a system.
• Potential failure modes can be deduced from the functions and parameters of each component.
The possible causes associated to each mode of failure considered need to be identified and described, in order to estimate their probability of occurrence, discover side effects and provide recommended corrective actions. A failure mode can have more than one cause.
Local effects are the failure mode consequences in the element of the system in consideration, while end effects are the impact produced in the whole system.
Another key concept is criticality. This is the degree of importance that is granted to a certain failure situation, which depends both on its probability of occurrence and the severity of its effects. In order
14 to evaluate it, it is required to identify elements that should be the subject of a more in-depth study in order to eliminate a risk, elements which require special attention during manufacturing or manipulation, special requirements to include in the market specification, and consider acceptance standards for the products of subcontractors, any special procedure and consider the most profitable application of the risk prevention.
Three elements therefore are relevant in the failure mode assessment:
• Occurrence, or sometimes termed likelihood, is a numerical subjective estimate of the likelihood that the cause of a failure mode will occur during the designed life. Usually, goes from 1 to 10, being 1 a remote failure possibility and 10 an inevitable failure.
Figure 9 Occurrence reference table example
• Severity is a numerical subjective estimate of how severe the end user will perceive the effect of a failure. The more severe, the higher the value will be. Follows the same scale as occurrence, with 1 being no impact of the failure and 10 injury danger.
Figure 10 Severity reference table example
• Detection is sometimes termed effectiveness. It is a numerical subjective estimate of the effectiveness of the controls to prevent or detect the cause or failure mode before the failure
15 reaches the customer. Again, follows the same scale of the other two indicators, being 1 certain detection and 10 not detectable failure.
Figure 11 Detection reference table example
If detectability is taken into account in addition to criticality, RPN is obtained. The Risk Priority Number values range from 1 (absolute best) to 1000 (absolute worst) and is a measure of the attention needed for each component of the system according to the FMCEA analysis.
FMECA is done accordingly to the standard UNE-EN 60812.
3.6 Fault Tree Analysis – FTA
The Fault Tree Analysis has for its object the identification and the analysis of the conditions and factors that cause or contribute to the appearance of a determined undesirable event, normally an event that affects notably the good operation of the system, its economy, security or other required feature.
The aim of an FTA is to identify the causes or combinations of causes that cause the superior event, determine whether a measure of the reliability of a system meets a specified requirement, demonstrate that the hypotheses made in other analysis are fulfilled, with respect to the independence of the systems and the non-relevance of the failures, determine the factor that most importantly affects a certain reliability characteristic, as well as the necessary modifications to improve that characteristic and identify common events or common cause failures.
To steps to follow for performing an FTA are the following:
1. Definition of the scope of the analysis.
2. Familiarization with the design, functions and operation of the system.
3. Definition of the superior or last event.
4. Construction of Fault Tree Analysis.
5. Analysis of the logic of FTA.
6. Reporting.
FTA is built in a top-to-bottom strategy, starting with the superior event and finishing with the basic events: superior event must be placed at the top and basic events are located at the bottom and end the FTA. Furthermore, security operating analysis must be documented to allow revision and
16 modification. The functional and physical limits should be clearly defined. Unity operation should not depend from any auxiliary function. No event will be related with other unity placed in a different place of FTA.
To evaluate the FTA, is necessary to identify events that can generate directly a system failure, and calculate its probability, estimate tolerance to system failure (ability to operate even after the occurrence of the fault or lower level event), verify the independence of failures of systems, subsystems or components and evaluate the data in order to locate critical components and failure mechanisms. Each event of the fault tree should be labelled uniquely. Cross reference should be easy to follow. There is only one superior event for each FTA and is the reason of the analysis. The same event in different locations of the FTA should carry the same label. Joins events are made by logic gates. These gates have only one output but can have multiples inputs. They are AND, OR and NOT gates.
In terms of logic analysis, there are three methods:
• Investigation: Revision of the structure of the fault tree, identification of common cause events and the search for independent branches.
• Boolean reduction: Is used for the evaluation of the effects of common cause events in fault trees in which the superior event does not depend on the moment or the order of occurrence of events.
• Method of minimum cuts: A cut is a group of events that, when occurring together, are the cause of the superior event. The minimum cut is the smallest one.
The purpose of numerical analysis is to provide a quantitative estimate of the probability of the occurrence of the superior event or a selected set of events.
If required, the FTA is done following the standard IEC 61025.
3.7 Reliability Block Diagram
Reliability block diagram (RBD) is a graphical analysis technique, which expresses the concerned system as connections of several components in accordance with their logical relation of reliability.
Series connections represent logic ‘‘and’’ of components, and parallel connections represent logic
‘‘or’’. [17]
The system proceeds from left to right, with several valid paths for successful operation. When a component fails, the corresponding connection is cut off. As components start to fail, system continues operating until no successful path is available from leftmost node to rightmost node. At that point the system has failed.
In Diagnostic Coverage (DC) the failure detection is rarely of 100%. Thus, the total dangerous failure is divided into detected failure and undetected failure, with failure rate λDD and λDU respectively:
𝜆𝐷= 𝜆𝐷𝐷+ 𝜆𝐷𝑈
Repair rates of the two types of failure are also separated, µDD for dangerous detected failure and µDU for dangerous undetected failure, as below:
𝜇𝐷𝐷= 1
𝑀𝑇𝑇𝑅 , 𝜇𝐷𝑈 = 1 𝑇1
2 + 𝑀𝑇𝑇𝑅
17 T1 refers to proof test interval and MTTR mean time to repair.
The structure function of a series structure is:
𝑅𝑆 = ∏ 𝜆𝑛
𝑛
𝑖=1
The structure function of a parallel structure is:
𝑅𝑆= 1 − ∏(1 − 𝜆𝑛)
𝑛
𝑖=1
With n referring to the number of subsystems.
If required, the RBD is done following the standard IEC 61078.
3.8 Rotating machine diagnosis
3.8.1 Gear Mesh FrequencyThe level of the tooth mesh frequency is dependent on the alignment of the shafts carrying the gears, and the load on the gear. A high peak at the gear mesh frequency does not necessarily indicate a problem, however an increase in amplitude without an increase in sidebands or harmonics suggests that tooth load has increased. [18][19]
Figure 12 Typical frequency spectrum of a geared system[20]
The gear mesh frequency is different for each gear assembly and appears in the frequency spectrum regardless the condition of the gears.
𝐺𝑀𝐹 = 𝑍 · 𝑅𝑃𝑀
In the field of digital signal processing, the sampling theorem is a fundamental bridge between continuous-time signals and discrete-time signals. “It establishes a sufficient condition for a sample rate that permits a discrete sequence of samples to capture all the information from a continuous- time signal of finite bandwidth”[21].
If a function contains no frequencies higher than “B” hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart[22].
A sufficient sample-rate is therefore anything larger than 2B samples per second. Equivalently, for a given sample rate, perfect reconstruction is guaranteed possible for a bandlimit B < f /2.
18 3.8.2 Bearing failure frequencies
Figure 13 General bearing breakdown[23]
Generally, bearing failures are not caused by the proper bearing, but what surrounds them. These problems generate abnormal vibration spectra. The most relevant are imbalance, misalignment, rotor instability, over or underload and mechanical looseness. Defective bearings that leave the manufacturer are very rare, and it is estimated that defective bearings contribute to only 2% of total failures [24]. Most of the problems are produced by dirt, shipping damage, storage or handling, wrong fitting when installing, wrong bearing chosen, improper lubrication practices, misalignment, bent shaft, imbalance, resonance and soft foot. A bearing suffering any of these conditions will be eventually damaged and destroyed.
A working bearing, ideally [25], has two distinct phases. In the first one, the bearing is operating in good conditions and the health indicator stays stable. While in the second phase, after a big change in the health indicator, the bearing starts its degradation entering an abnormal health condition, degrading exponentially over time until the health indicator reaches a user-defined failure threshold.
The health indicators are extracted directly form the vibration signal pre-processing, characterizing them using statistical parameters.
While temperature signals are only sensitive to severe bearing failures, vibration and acoustic signals are commonly used for fault diagnosis and prognostic analysis of bearings [25]. The deterioration of each element will generate different failure frequencies in the frequency spectrum which will allow to easily identify them. These frequencies are:
• BPFO (Ball Pass Frequency Outer), also known as outer race failure frequency, which corresponds with the number of rollers or balls that pass through a point of the outer race each time the shaft makes a complete turn.
𝐵𝑃𝐹𝑂 =𝑛 · 𝑁𝑏
2 · (1 +𝑑
𝐷cos (∅))
• BPFI (Ball Pass Frequency Inner), also known as inner race failure frequency, which is the number of rollers or balls that pass through a point of the inner race each time the shaft makes a complete turn.
𝐵𝑃𝐹𝐼 =𝑛 · 𝑁𝑏
2 · (1 −𝑑
𝐷cos (∅))
19
• BSF (Ball Spin Frequency), also known as rolling element failure frequency, which is the numbers of turns that a bearing rolling element or ball makes each time that the shaft makes a complete turn.
𝐵𝑆𝐹 = 𝐷
2𝑑𝑛 · (1 − (𝑑 𝐷)
2
(cos(∅))2)
• FTF (Fundamental Train Frequency), also known as cage failure frequency, which is the number of turns that makes the bearing cage each time the shaft makes a complete turn.
𝐹𝑇𝐹 = 𝑛
2𝑑· (1 −𝑑
𝐷cos (∅))
20
4 Method
In order to carry out this project, five phases are defined:
1. A study of the state of the art and current system is done. To fulfil this, a study of the taxonomy and ontology of the system is required, identifying all the elements to be included in the project.
2. A Failure Mode Effects and Criticality Analysis (FMECA) is done in order to detect failure modes for the elements under study, and a Fault Tree Analysis (FTA) to identify the relation between them. FMECA is done accordingly to the standard UNE-EN 60812.
3. Once the failure modes are specified, it’s necessary to evaluate different approach architectures, for which knowledge of all the work values are required, for example, operating frequencies or motor speed.
4. Sensors location can be crucial for obtaining reliable signals. Depending on the type of industry and industrial equipment, perfect sensor position may vary, so a viable sensor implementation strategy is also discussed. The criteria for sensor selection will be based on the actual machine and the acquisition system topology.
5. Once the whole system is ready to read data from the monitorization, processing techniques are needed in order to complete data analysis.
To achieve a higher grade of knowledge, the study will base on a project done by Tecnalia in a food industry, and it will take place both in the installations of Tecnalia R&I and the industrial company:
1. Element identification: taxonomy and ontology. The study of the whole system is done. This phase consists in the breakdown of the system into individual elements and how they are related. The taxonomy is done following the standard ISO 14224.
2. Failure mode identification, through FMECA and FTA, taking advantage from data obtained from the machine. Identification of failure modes and study of the effects and criticality on the system. Information about failures to complete the FMECA is discussed with the industrial company, who has exact failure rates data. Specifically, the contact person in the company is the maintenance manager, who can have access to maintenance data and operation information. Hidden failure modes also are identified if necessary.
3. Frequency analysis has a big role in fault detection. By studying the operating frequencies of the components makes possible the implementation of sensors and the study of the monitored data in order to compare it to the standard operating footprint. This operating frequencies, which are RPM of the different stages, are obtained from the motor and gearbox plates accessed through the maintenance manager.
4. Data analysis proposal, in two approaches: the first one determines the capabilities of the predictive maintenance system; the second one after all the sensors and the system are implemented, propose stationary (FFT) and non-stationary (STFT) modes, which will be reflected with test data.
5. Instrumentation: evaluation of the available options and comparison of type of sensors and their position. The “best” monitoring system is selected evaluating both prize and adaptability of the system.
6. Data analysis, feature extraction. Presenting different processing techniques, which one fits best the online condition monitoring.
7. Integration with existing systems. If viable, integrate the developed maintenance system into the company existing structure.
21
5 Results and discussion
5.1 Technical Breakdown: System enumeration
In order to achieve a global view of the system, a taxonomy or technical breakdown of the system is done. To do so, the system is divided into three overall subsystems, which will contain the different components.
These three systems are:
• Electric motor
• Input shaft
• Fixed shaft
As shown in figure 14, the different parts of the process can be identified even inside the same machine.
Figure 14 System taxonomy
5.2 FMECA: Potential failure modes identification
The next step is the identification of failure modes. This is achieved through FMECA, taking advantage from data obtained from the machine.
From the system division done in the taxonomy and supporting the decisions with both the machine owner and the technical knowledge, the failure modes can be assessed. After the FMECA is done, an overview of the most critical failure modes can be obtained, as displayed in the following figure:
22
Figure 15 Top 10 RPN from FMECA
In the previous graph, the Top 10 RPN values are displayed. This is a measure of which modes, and therefore, which components, require more maintenance attention.
Mode ID Failure Mode Mode ID Failure Mode
50 Input Shaft cracked/fractured 52 Bearing (6303) loss or deterioration of lubrication 77 Fixed Shaft cracked/fractured 64 Bearing (61811-2RS1) loss or
deterioration of lubrication 85 Fixed Shaft pinion worn out 89 Bush brinelling
76 Input Shaft pinion worn out 59 Bearing (6201) contamination 58 Bearing (6201) loss or
deterioration of lubrication
53 Bearing (6303) contamination
Figure 16 Top 10 RPN breakdown
This result makes the transmission system the most critical part of the system.
From the FMECA, the most important causes of failure can also be extracted, as displays figure 17:
23
Figure 17 Top 10 failure mode causes from FMECA
Furthermore, severity can be assessed too, by taking into account the severity distribution of the analysis done. In the figure, the different severity rates of the system are summarized. With the distribution being “red” severity rate 2, “yellow” severity rate 5, “green” severity rate 6, “blue”
severity rate 7, and “purple” severity rate 8; gives an overall idea of a medium severity system. Severity distribution is shown in figure 18.
Figure 18 FMECA severity distribution
The complete FMECA can be found in Appendix.
24
5.3 Frequency analysis
The study of the operating frequencies is confined to the elements identified by the FMECA as the most critical elements. This is the gears and the bearings.
5.3.1 Gear mesh frequency
Depending on the rotational speed of the electric motor, GMF will variate. In order to make sure that all the requirements, the worst-case scenario is studied, which is at maximum speed, that is 9000 rpm.
𝐺𝑀𝐹 = 𝑍 · 𝑅𝑃𝑀 = 27 ·9000 𝑟𝑝𝑚
60 𝑠 = 4050 𝐻𝑧
Gear Z m di (mm) RPM GMF (Hz)
Gear 1 (input) 27 1,5 18 9000 4050
Gear 2 67 1,5 57 3626,87 4050
Figure 19 Gear characteristics
5.3.2 Bearing frequencies
Each bearing has its own frequency features, so the bearing frequencies need to be calculated for all the different bearings in the system.
Bearing BPFO BPFI BSF FTF
6201 381, 818 668, 182 254,545 54,545
6303 381,757 668,243 254,419 54,537
61811-2RS1 643,019 747,281 400,761 27,957
Figure 20 Bearing frequencies of the system
Considering all the working frequencies, in order to size the acquisition system, the highest frequency gives the threshold. In this case, GMF is the highest frequency, and in order to see 3 times the GMF, which will be the third harmonic, and applying sampling theorem, at least a 13 kHz accelerometer sensor is required, and a 26 kHz sampling frequency is needed to rebuild the signal.
5.4 Acquisition system: Identification and monitoring of failures
Reliability prediction, traditionally, is based on failure times data. Nonce, maintenance actions start in terms of a reliability index. Usually, preventive maintenance strategies don’t consider the current health state of the equipment and base the replacement in time standards. On the other hand, a maintenance strategy developed taking into account the current degradation and evolution of the equipment is referred as Condition Based Maintenance (CBM). The basic principle of this technique is to use online sensing tools to obtain the current product degradation data and minimize the system downtime by balancing the risk of failure and achievable profits.
Techniques for condition monitoring are mainly related with vibrations and acoustic techniques, and some are centred on failures detection. But the real non-intrusive aim is to develop an integrated system for health assessment of manufacturing machines based on the smart analysis of machine operational conditions. Focusing on current signals of the drives, power consumption of the motors plus available information of the process through the controller, all of them easily accessible.
The diagnosis approach is divided into two main systems. The first one is the extruder head itself, and the second one is the conveyor belt motors.
As pointed out in the taxonomy and FMECA studies, the extruder head consists of 11 components.
Failure will most likely come from material stuck, but time degradation must also be supervised. The
25 diagnosis of this system will be done by analysing frequency spectrum of a triaxial accelerometer on the extruder head mechanism and frequency spectrum of the current of the motor powering the system. Additionally, temperature measurements will support the diagnosis system.
In the other hand, in the case of the conveyor belts, the diagnosis will only consist in a power monitoring, with the aim of detecting when the belt of the conveyor is about to collapse.
5.4.1 System topology
There are 3 decision system topologies:
1. Fully centralized decision system:
Each asset communicates sensed physical variables or features to Maintenance Center. The Maintenance Center computes health indicators for all assets and makes maintenance decision for all assets by optimizing overall cost function.
2. Semi-centralized decision system:
Each asset communicates only its health indicator to Maintenance Center. The Maintenance Center makes maintenance decisions for all assets by optimizing overall cost function.
3. Fully decentralized decision system:
Decision logic at local level, assets can share information.
From the three decision system topologies, the one that fits best for the desired operation is the semi- centralized decision system.
Figure 21 System topology
26 This will consist of a computing element, which will be called Edge Computing Unit, or simply Edge, that will simultaneously gather data for processing and, if needed, generate alarm and health features, and at the same time will send, with a stablished periodicity, raw data for its analysis by an expert.
The Edge will count with Datalogger boards, one for each sensor, that will be in charge of the acquisition of signals.
Furthermore, the Edge will need to access Internet, for what all the necessary security protocols will be followed.
Context info, as operating status, time of operation, etc. will be accessed from the company MES, the SCADA/PLC of the line or from the system CMMS, and the data analytics environment will be the company own cloud environment where data analysis will be done.
5.4.1.1 Pilot implementation
In order to develop a first pilot or demonstrator of the system as “lean” as possible some limitations have been made to the general scope of the system:
On the one hand, due to Internet security certifications, the device that will supervise and allow internet connection and, in extend, data transmission and analysis, will be the eWon Cosy. This is a secure industrial router and firewall that allows connection for example, to a database, and parallelly permits a secure VPN connection in order to configure the system remotely. This way, the expert team will have access to the obtained data and will enable the learning process, making decisions if necessary.
Besides, due to the complexity of achieving real-time IT systems connectivity, it has been decided that for pilot purposes, information related to extruder head changes and maintenance events will be shared off-line (via excel spreadsheets) between the company and Tecnalia, the analytics team.
Real time connectivity to the extrusion line PLC is also discarded for the time being.
Finally, although in a future the company expects to develop its own data analytics cloud environment where processing modules can be carried out, for the purpose of accelerating the pilot deployment this processing will be carried out at a server on Tecnalia premises.
27
Figure 22 System connectivity
5.4.2 Sensor selection 5.4.2.1 Accelerometer
As mentioned in section 4.5, the frequencies to take into account for sensor selection are the rotation speed, the gear frequency and the bearings frequencies. Since, for analysis, three times the Gear Mesh Frequency is important (3 x GMF), that makes the highest frequency 12150 Hz. To fulfil this requirement, a sensor of more than 12.5 kHz resolution before 3 dB attenuation is required.
For this purpose, the sensor PCB IMI 639A91 is selected.
Figure 23 PCB IMI 639A91
Consists of an industrial grade triaxial accelerometer sensor, with frequency response on all three axes up to 13 kHz. The mounting is screwed. The three-axis characteristic will allow to detect anomalies in the horizontal plane, due to motor rotation, and for the bearings both in the vertical and axial
28 (longitudinal) plane, due to being deep groove bearings. Gear failures can also be detected easily given that all directions are supervised.
5.4.2.2 Temperature
Temperature is not a signal to process, only will allow failure correlation and alarm going-off. Thus, it is not required to obtain high frequency signal from temperature. To ease extruder caliber changing, the most viable approach is to use a contactless sensor, mounted in a non-changing surface.
For this purpose, the optic sensor Optris CSmicro is selected, due to its range, from -50 to 1030 ºC.
Furthermore, it has the controlling electronics embedded in the cable, which makes it very easy to handle.
Figure 24 Otris CSmicro with mounting bracket ACCTFB (left) and schematic with dimensions (right)
Additionally, a mounting bracket will be needed in order to make sure it is always measuring the same point. The location can be any that allows it to point to the extruder head surface.
Sensor will be mounted with the help of the adjustable in one axis mounting bracket ACCTFB.
Furthermore, it can be placed over a hinge for easy head changing. This hinge will need to be manufactured specifically for the application.
5.4.2.3 MCSA sensor: Current transformer
In order to perform MCSA, it is required a current transformer with very wide frequency bandwidth.
Since normal current transformers are for one fixed frequency (0-DC), it is an uncommon component.
The sensor LEM LTC 200-S is selected, because it has a bandwidth from 50 Hz to 100 kHz, that fits the required bandwidth for the application, which is around 5-10 kHz. Additionally, supply voltage is +/- 15 V, which requires a specific power supply, which will be a switched-mode power supply of +/- 15 V, the Tracopower TXL 035-1515D. The sensor will be placed on the output of the motor controller.
Figure 25 LEM LTC 200-S current transformer and Tracopower TXL 035-1515D power supply
29 5.4.2.4 Power monitoring sensor: Current transformer
For the conveyor belt motor, much lower frequency will be used, so a simpler transformer can be used. The Circutor FT MC3 is selected with its main advantages being the small size and that all 3 phases are on the same capsule. Voltage will be directly connected to the datalogger module.
Figure 26 Circutor FT MC3 (left) and example of real application (right)
5.4.3 Implementation architecture 5.4.3.1 Edge computing Unit (Industrial PC)
The main requirements for this unit are a good processing capacity and connectivity.
The Beckhoff CX5140-0155 is and Industrial PC that fulfils both requirements. It´s a DIN-rail- mountable, fan-less Embedded PC, equipped with Intel Atom multi-core processor.
Figure 27 Beckhoff CX5140 connexion
5.4.3.2 Accelerometer datalogger module
In signal sampling, the frequencies are considered again, taking into account the sampling theorem.
That is, with a 13 kHz accelerometer, sampling frequency needs to be at least 26 kHz. The module Beckhoff EL3632 is capable of 50 ksamples per second with 16-bit resolution. This is a 2-channel module; hence 2 modules will be required for the 3 signals.
30
Figure 28 Beckhoff EL3632 connexion
5.4.3.3 Temperature datalogger module
Since, as already stated, temperature is not a critical signal, in terms of speed sampling, a basic voltage input module will be used. This is the Beckhoff EL3312.
Figure 29 Beckhoff EL3161 connexion
5.4.3.4 High frequency current datalogger module
Current will be measured at high frequency to allow MCSA. Anyhow, the frequency will be lower than the vibration signals, therefore, a similar module to the accelerometer datalogger regarding frequency is required, with 0-1 A input range for current transformer output. The module Beckhoff EL3773 fits the requirements.
Figure 30 Beckhoff EL3773 connexion
31 5.4.3.5 Conveyor motor current datalogger module
For the conveyor motor current, a power monitoring module is enough to supervise the operation of the component. This is a 3 voltage input and 3 current input, from 0 to 1 A for the output of the current transformer. For this purpose, the module EL3443 is selected.
Figure 31 Beckhoff EL3443 connexion
32 5.4.4 Connexion
Figure 32 Acquisition system connexion
-15V +15V
33
5.5 Data analysis plan
As already explained, a semi-centralized communicates only its health indicator to the Maintenance Center, which then makes maintenance decisions for all assets. In this case, a special variant of system is implemented, where the on-site element pre-process data and as well sends raw data for analysis in the Maintenance Center.
5.5.1 Strategy
The approach to follow when it comes to data processing and transmission will be the following:
• Three cycles of operation per minute will be obtained. The time will depend of each extruder head cycle time.
• The pre-processing algorithms consist in the calculation of the FFT of each signal (3- axis accelerometer and 1 phase extruder motor current) and additionally the energy of each FFT. Conveyor belt motor current and temperature will be acquired at a much lower frequency, so they both can be sent without pre-processing.
• This data will be then sent in packages every ten minutes.
• Furthermore, once per day, the complete raw data of the three cycles will be also sent.
Therefore, the Maintenance Center can feedback the pre-processing algorithms and complete data is available for further analysis if necessary. Also, raw data at the moment of start of operation is susceptible of being sent, due to being non-stationary state.
5.5.2 Pre-processing techniques
The pre-processing techniques that are going to be employed are the Fast Fourier Transform and the energy of that FFT.
FFT is suitable for characterization of stationary signals. However, it is not suitable for signals with transitory characteristics. This is why 3 cycles in stationary state will be analysed.
Additionally, the energy of the FFT will be calculated in a point where the function has a maximum and its surroundings. That is, the maximum points of the FFT with a neighbourhood of +/- 2 dBs will be integrated.
5.5.3 Graphical results
Below the graphical results of two operating machines will be discussed. Top graph shows the sensor data for a low use machine, while bottom one shows the data for a worn (in comparison with the first one) machine.
The first figure shows the overall vibration spectrum for both machines.
34
Figure 33 Overall vibration spectrum
The next figure shows detailed low frequencies, around Rotating Frequency (Fr).
Figure 34 Rotating frequency zoom in
Following, a detail of the GMF x2 is shown. Is important to notice the difference in amplitude between both graphs, as well as the GMF x2 +/- Fr growing.
Figure 35 GMF x2 zoom
35 The same is done with the 3x GMF. It can be seen how in the low use machine there is no 3x GMF and on the “worn” machine 3x is more visible.
Figure 36 GMF x3 zoom in
Finally, a comparison of the current spectrum is done. It can be seen a zoom around Line Frequency (FL). Again, the second graph shows growing peaks in FL +/- Fr. Since there is a gearbox, both rotating frequencies can be seen.
Figure 37 Current spectrum detail
36
6 Conclusion
Firstly, results are based just in one machine. This has direct impact in objective 4, since a bigger number of machines can influence the decision taking of the topology or strategy employed.
Furthermore, also impacts objective 6, because in early stages of the project is difficult to randomly find a failure, due to all equipment working ok or at least nearly ok, because industries don’t use to keep failed equipment working. Additionally, costs of machine stop are not available.
After the development of this project, the study of the failure modes has been done and also a draw out the critical elements of the system that have been focused. Following these critical elements, the condition monitoring system was designed and effectively tested to detect them. The study done to determine which technique will be more appropriate concluded that a combination of both current and vibration sensors provides the maximum amount of information about the system. Additionally, due to infrastructure limitations, that information is extracted to an external server where it is stored and processed. Applying this solution aims to be proven useful in both improving downtimes and preventing failures, thus improving availability, although it hasn´t been tested so far yet.
Lastly, it can be discussed whether the project should have been aimed in any other way, for instance employing wireless sensors, but most importantly, if it would have been possible to perform the data processing in the edge device, which would have been a much more efficient data transfer technique.
Additionally, if costs of failures and downtimes would have been available, a Life Cycle Cost analysis could have been done, where the real impact of having or not having a Condition Monitoring System would be shown.
Future line of works that can be draw from this project are the LCC analysis just mentioned, improving processing algorithms to automate data processing, memory ampliation for continuous monitoring, application of machine learning and neural networks to the system, and obviously, the monitoring of more than one machine.
37
7 References
[1] M. Kajko-Mattsson, R. Karim, and A. Mirijamdotter, ‘Fundamentals of the eMaintenance Concept’, p. 8, 2010.
[2] J. Lee, J. Ni, D. Djurdjanovic, H. Qiu, and H. Liao, ‘Intelligent prognostics tools and e-maintenance’, Comput. Ind., vol. 57, no. 6, pp. 476–489, Aug. 2006, doi: 10.1016/j.compind.2006.02.014.
[3] R. Smith, ‘Regulation (EC) No 764/2008 of the European Parliament and of the Council’, in Core EU Legislation, London: Macmillan Education UK, 2015, pp. 183–186.
[4] V. Atamuradov, K. Medjaher, P. Dersin, B. Lamoureux, and N. Zerhouni, ‘Prognostics and Health Management for Maintenance Practitioners - Review, Implementation and Tools Evaluation’, p.
32.
[5] Galar, Diego, Sandborn, Peter, and Kumar, Uday, Maintenance costs and life cycle cost analysis.
2017.
[6] I. B. Imaz, ‘Acquisition and processing of new data sources for improved condition monitoring of mechanical systems’, p. 200.
[7] A. K. S. Jardine, D. Lin, and D. Banjevic, ‘A review on machinery diagnostics and prognostics implementing condition-based maintenance’, Mech. Syst. Signal Process., vol. 20, no. 7, pp. 1483–
1510, Oct. 2006, doi: 10.1016/j.ymssp.2005.09.012.
[8] ‘An308 Current Sensor Jan09 | Electric Current | Transformer’, Scribd.
https://www.scribd.com/document/211243840/An308-Current-Sensor-Jan09 (accessed Sep. 21, 2020).
[9] M. Messaoudi, L. Sbita, and M. N. Abdelkrim, ‘Faults Detection in Induction Motor via Stator Current Spectrum Analysis’, p. 7.
[10] S. Singh, D. Galar, D. Baglee, and S.-E. Björling, ‘Self-maintenance techniques: a smart approach towards self-maintenance system’, Int. J. Syst. Assur. Eng. Manag., vol. 5, no. 1, pp. 75–83, Mar.
2014, doi: 10.1007/s13198-013-0200-7.
[11] V. Simon, C.-A. Johansson, and D. Galar, ‘Aggregation of Electric Current Consumption Features to Extract Maintenance KPIs’, Manag. Syst. Prod. Eng., vol. 25, no. 3, pp. 183–190, Sep. 2017, doi:
10.1515/mspe-2017-0027.
[12] S. Ferreiro, E. Konde, S. Fernández, and A. Prado, ‘INDUSTRY 4.0: Predictive Intelligent Maintenance for Production Equipment’, p. 9, 2016.
[13] M. Messaoudi and L. Sbita, ‘Multiple Faults Diagnosis in Induction Motor Using the MCSA Method’, p. 7.
[14] Y. Lei, N. Li, L. Guo, N. Li, T. Yan, and J. Lin, ‘Machinery health prognostics: A systematic review from data acquisition to RUL prediction’, Mech. Syst. Signal Process., vol. 104, pp. 799–834, May 2018, doi: 10.1016/j.ymssp.2017.11.016.
[15] K. T. Huynh, Y. Langeron, and A. Grall, ‘Degradation Modeling and RUL Estimation of Deteriorating Systems in S-Plane’, IFAC-Pap., vol. 50, no. 1, pp. 12249–12254, Jul. 2017, doi:
10.1016/j.ifacol.2017.08.2036.
[16] P. Do Van, E. Levrat, A. Voisin, and B. Iung, ‘Remaining useful life (RUL) based maintenance decision making for deteriorating systems’, IFAC Proc. Vol., vol. 45, no. 31, pp. 66–72, 2012, doi:
10.3182/20121122-2-ES-4026.00029.
[17] H. Guo and X. Yang, ‘A simple reliability block diagram method for safety integrity verification’, Reliab. Eng. Syst. Saf., vol. 92, no. 9, pp. 1267–1273, Sep. 2007, doi: 10.1016/j.ress.2006.08.002.
[18] W. D. Mark, A. C. Isaacson, and M. E. Wagner, ‘Transmission-error frequency-domain-behavior of failing gears’, Mech. Syst. Signal Process., vol. 115, pp. 102–119, Jan. 2019, doi:
10.1016/j.ymssp.2018.05.036.
[19] A. Saxena, M. Chouksey, and A. Parey, ‘Effect of mesh stiffness of healthy and cracked gear tooth on modal and frequency response characteristics of geared rotor system’, Mech. Mach. Theory, vol. 107, pp. 261–273, Jan. 2017, doi: 10.1016/j.mechmachtheory.2016.10.006.
38
[20] ‘Gears - Hunting Tooth Frequency’.
http://www.vibrationschool.com/mans/SpecInter/SpecInter50.htm (accessed Sep. 18, 2020).
[21] ‘The “Sound” of Performance Monitoring, Part 2: Aliasing’, Aternity, Jul. 17, 2018.
https://www.aternity.com/blogs/the-sound-of-performance-data-part-2/ (accessed Sep. 21, 2020).
[22] ‘Sigproc Selfstudy | Telecommunications | Electronic Engineering’, Scribd.
https://www.scribd.com/document/234168119/Sigproc-Selfstudy (accessed Sep. 21, 2020).
[23] ‘Rolling element bearing components and failing frequencies’, Power-MI, Jun. 08, 2017.
https://power-mi.com/content/rolling-element-bearing-components-and-failing-frequencies (accessed Sep. 18, 2020).
[24] R. Smith and R. K. Mobley, Rules of Thumb for Maintenance and Reliability. 2008.
[25] D. Wang, K.-L. Tsui, and Q. Miao, ‘Prognostics and Health Management: A Review of Vibration Based Bearing and Gear Health Indicators’, IEEE Access, vol. 6, pp. 665–676, 2018, doi:
10.1109/ACCESS.2017.2774261.
39
8 Appendix
8.1 FMECA
40
41
42
8.2 FTA
43
