Reductions in Energy Consumption through Process Optimisation and Variable Production

Bachelor Degr ee Pr oject

7

Vår logotyp ska göra att man snabbt känner igen Högskolan i Skövde i alla sammanhang där vi syns.

Logotypen är en fast graf sk kombination med noggrant avvägda proportioner.

Därför får den inte ändras eller förvanskas på något sätt. Rätt använd gör den Högskolan i Skövde tydlig som presentatör eller avsändare.

FRIZON

Runt logotypen f nns ett reserverat om- råde motsvarande bokstaven H:s versal- höjd i ordbilden. I detta område skall inga andra graf ska element placeras.

MINSTA STORLEK

För att logotypen ska vara tydlig bör den aldrig vara mindre än 27 mm bred.

LOGOTYPENS DELAR

Logotypen består av två graf ska element:symbol och ordbild.

Elementen används aldrig separat.

Logotypen används alltid i sin helhet.

Symbolerna i vapnet bygger på Västergöt- lands lejon för att framhäva vår geograf ska härkomst samt boken och trädet kopplad till vår akademiska verksamhet att ut- bilda och skapa kunskap. Logotypen har även årtalet 1977 då verksamhet startades vid Högskolan i Skövde.

LOGOTYP

Symbol

Ordbild

Reductions in Energy Consump- tion through Process Optimisa- tion and Variable Production

Bachelor Degree Project in Automation Engineering 30 ECTS

Spring term 2017 Idir E

Itsaso M

Supervisor: Amos N

Examiner: Sunith B

Högskolan i Skövde

Abstract

School of Engineering Science

Automation Engineering

Reductions in Energy Consumption through Process Optimisation and Variable Production

by Idir E

and Itsaso M

Energy efficiency is becoming an important objective for modern manufacturing in- dustry. The aim of this work is to improve energy efficiency of an automated system.

Since a majority of production processes are limited by an external bottleneck, the hypothesis of this work is that reducing the processing rate of the restricted pro- cesses can lead to saving in energy and resources. A methodology based on optimi- sation at process, cell and line levels is developed and evaluated over different sce- narios. The developed methodology is then applied to a simulated production cell to study its efficacy quantitatively. In this particular case, the proposed approach yields a decrease in energy consumption of 49% at maximum production capacity.

This decrease can be greater if there is an external factor such as low demand or

another stage in the production line.

Contents

Certificate of Authenticity iii

Abstract v

1 Introduction 1

1.1 Background . . . . 1

1.2 Aim and Objectives . . . . 1

1.3 Outline . . . . 2

2 Theoretical Framework and Literature Review 3 2.1 Energy Consumption for Common Actuators and Optimisation Via- bility . . . . 3

2.1.1 Electric Motors . . . . 3

Energy Efficiency in Motors and Drives . . . . 3

2.1.2 Robots . . . . 4

2.1.3 Pneumatic Systems and Actuators . . . . 5

2.2 Effects on Quality and Application Scenarios . . . . 5

2.2.1 Material Handling . . . . 6

2.2.2 Machining . . . . 6

2.2.3 Assembling . . . . 6

2.2.4 Welding . . . . 7

2.2.5 Painting, Coating and Similar . . . . 7

2.3 Modelling and Simulation . . . . 8

2.3.1 Single Process Simulation . . . . 8

Modelica . . . . 8

2.3.2 Cell Simulation . . . . 8

Simumatik3D . . . . 8

2.3.3 Production Line Simulation . . . . 9

3 Methods 11 3.1 Process Optimisation . . . 12

3.2 Cell Optimisation . . . 13

3.3 Line Optimisation . . . 14

4 Case Study 15 4.1 Conveyor Optimisation . . . 15

4.1.1 Conveyor Model . . . 15

4.1.2 Motor Parametrisation . . . 17

4.1.3 Load Parametrisation . . . 18

4.1.4 Optimisation . . . 18

4.2 Cell Optimisation . . . 19

4.2.1 Optimisation Algorithm . . . 19

4.2.2 Optimisation Results . . . 20

viii

5 Conclusions and Discussion 23

5.1 Conclusions . . . 23 5.2 Discussion and Future Work . . . 23 5.3 Sustainability Perspective . . . 24

Bibliography 25

List of Figures

2.1 Potential savings from VSD . . . . 4

2.2 Speed energy ratio for IR . . . . 5

3.1 Control tree . . . 11

3.2 Overview of the optimisation process . . . 11

3.3 Pareto front example . . . 12

3.4 Metaheuristics classification . . . 13

4.1 Cell distribution . . . 16

4.2 Conveyor model in Modelica . . . 16

4.3 Equivalent circuit of an induction machine . . . 17

4.4 Pareto front of conveyor . . . 18

4.5 Diagram of cell input and outputs . . . 19

4.6 Pareto front for the cell . . . 20

4.7 Complete optimisation diagram . . . 21

List of Abbreviations

AC Alternating Current DC Direct Current EM Electric Motors

FMI Functional Mock-up Interface FMU Functional Mock-up Unit IR Industrial Robots

MOSHC-R Multi-Objective Stochastic Hill Climbing with Restart NSGA-II Non-dominated Sorting Genetic Algorithm II

OPC OLE for Process Controll

PLC Programable Logic Controller

PWM Pulse Width Modulation

VFD Variable Frequency Drive

VSD Variable Speed Drive

Chapter 1

Introduction

1.1 Background

From the very invention of the production line, one of the main goals of production engineering has been to increase the output yield of manufacturing systems.

Increasing the production rates of a system has many advantages, such as an improvement in the profitability of the facilities, lowering running and operational costs, maximising the use of the company’s resources, supplying a greater demand, taking a greater portion of the market, et cetera.

This line of progress has led to the current scenario, where most manufacturing systems are built and optimised to produce at the highest rate possible, capable of keeping up with peak demands and even more. According to a US Department of Energy study, 44% of motors in industrial facilities operate at 40% or less of their full load capacity (Saidur, 2009). As common as oversized systems are, there seems no to be a clear strategy when the demand is lower than the production capacity for which the line was designed.

Whenever there is a limiting factor in the production flow, previous machines and processes have to limit their production or store the surplus. This limiting factor can be of any kind, from a low demand to the production bottleneck or a repairing machine, and the solution goes from placing great-enough buffers where this limit- ing factors can occur to intermittently activating the processes to match the needed production rate. Machines are, therefore, producing at maximum capacity or com- pletely stopped.

Is in this context lays the niche for this work. The proposed strategy is to add flexibility to previous rigid systems by optimising every process in the line, mak- ing it capable of adapting to the desired or needed production capacity. The main hypotheses to be studied in this thesis are that systems that operate at a lower pro- duction rate are more energy efficient, require less maintenance and have a longer live span.

If favourable, the results of this thesis may apply to an extended part of the man- ufacturing sector, leading present and future industries to a more sustainable and profitable future.

1.2 Aim and Objectives

The aim of this work is, therefore, to study how to improve the efficiency of an automated system through adapting it to the needed production capacity, as well as analysing the viability of the proposed technique in various common scenarios.

The concrete objectives will be to study precedents in the field and set the poten-

tial benefits of this idea. Several kinds of actuators, such as electric and pneumatic,

2 Chapter 1. Introduction

need to be analysed to find out if they are optimisable. Probably not every process responds well to the hypothesis presented above, so is necessary to determine under what circumstances and in which scenarios efficiency and production capacity are not conflicting objectives, and therefore the optimisation process would fail to find a frontier between them.

The presented technique needs to be applicable in the industrial sector, so the proposed approach to the line optimisation will be delineated focusing on ease of implementation and simplicity. Finally, to validate the proposals, the method will be applied to a case of study in which to find possible problems with the implemen- tation and to measure improvements. This case of study will consist in a simulation of a manufacturing cell, modelled to serve as an optimisation tool and as a real-time simulation to test the results.

Energy consumption is a major concern in the industry, and a vast quantity of studies focus on how to improve many operations in various scenarios. This work does not intend to optimise the specifics of every process, and it will be assumed that the most significant aspects for efficiency have already been contemplated.

1.3 Outline

After this introduction, the viability of this technology will be studied in various

activities and for different actuators in Chapter 2, as well as an overview of the tools

used in the case of study. In Chapter 3 an implementation method is proposed. In

Chapter 4 is developed a case of study to analyse the viability and problems in a real

case scenario. Finally, the results of the tests are analysed and a global discussion of

the work is carried out in Chapter 5.

Chapter 2

Theoretical Framework and Literature Review

2.1 Energy Consumption for Common Actuators and Opti- misation Viability

2.1.1 Electric Motors

In a production line, to move products from one station to the next one conveyor belts are used. Electric motors (EM) are the main way to drive these conveyors and can be powered by AC or DC.

DC motors are commonly used in industrial applications because the speed- torque relationship is easier to adjust than AC motors. Besides, they are employed in applications where the power source is DC (e.g. automobile motors).

AC motors are separated into two types: Induction and synchronous. Induction machines are simple and usually cheap, and they are used in applications where the speed does not need to be varied. Synchronous motors are more complex and mainly utilised in the major industrial applications (Beaty and Kirtley, 1998).

Energy Efficiency in Motors and Drives

Electric motors are the most significant load in the industry, and they sum for about 70 % of the consumed energy in the European Union (EU) (Almeida, Ferreira, and Both, 2005). In consequence, a lot of studies have emerged to investigate how to reduce the overall power consumption of these.

In industry, EM has been designed and implemented to operate at maximum power when they are needed and to stop completely when not. This is not, by far, the best strategy to drive them and since the vast majority of EM in the industry are significantly oversized and significant energy savings can be made controlling the velocity and torque of the motor (Saidur, 2009).

Variable speed drives (VSDs) are used to regulate the speed of electric motors.

VSD is the best technology to save potential energy of the motors. It is an electronic

power converter that generates a multi-phase, variable frequency output that con-

trols the motors speed torque and mechanical power output (Abdelaziz, Saidur, and

Mekhilef, 2011). This technology can be used with both DC and AC machines. In

the case of AC motors, synchronous and induction motors are coupled to inverters

that generate variable frequency (the speed is proportional to the frequency) (Beaty

and Kirtley, 1998). The most used VSD is the variable frequency drive (VFD).Which

using power electronic components, it controls the motor speed by changing the in-

put frequency. Most used VFD uses pulse width modulation (PWM).(Saidur et al.,

2012)

4 Chapter 2. Theoretical Framework and Literature Review

10 15 20 25 30 35 40 45 50 55 60

20 40 60 80

Average percent speed reduciton [%]

Potential ener gy savings [%]

F

2.1: Potential savings from VSD (Motor energy saving)

The energy consumption of a motor using VSD depends on the energy consump- tion without using VSD and speed reduction ratio (SR). The SR is maximum engine speed divided by its rated speed. So to make a mathematical calculation of energy consumption Equation 2.1 is used. (Abdelaziz, Saidur, and Mekhilef, 2011)

P = P

(1 − SR)

(2.1)

where P is the energy consumption of the motor using VSD, P

is the power consumption of the motor without speed control, and SR is the speed reduction ratio.

Figure 2.1 shows the potential energy savings associated with the speed reduc- tion using VSD in industrial motors. As it can be seen, the potential energy savings increases with the decrease in velocity.

2.1.2 Robots

The use of industrial robots (IR) has increased in the last decade in the mechatronic industries due to their great flexibility. Their energy consumption adds up to ap- proximately 8% of the total electricity consumed in production processes. Because of this, there have been a significant number of recent investigations to reduce the con- sumption of IR at the process level, mainly centred on optimising the path, breaking times and smoothness of the movement (Paryanto et al., 2015).

Given the number of independent investigations on this topic and already ex- isting commercial solution, we will assume that the robot path has been previously optimised and will focus on adjusting the parameter of the pre-programmed robot.

Despite that, the work of Riazi et al. (2015) is very relevant in our context, since it is focused on reducing the energy consumption of IR without altering their path, only optimising the acceleration along it. They have developed a quick and easy to implement algorithm that minimises the square acceleration of every joint with some constraints, among which is the total time of the movement. This algorithm can be easily modified to adapt to any given production rate, enabling the robots to accommodate to the slowed down production line efficiently.

If the previous algorithm is not implemented, a proportional reduction in the

speed of the robot movements can achieve by itself a significative reduction in en-

ergy consumption, as seen on Figure 2.2.

0 20 40 60 80 100 2

3 4 5

Robot program execution time [s]

Ener gy consumption

F

2.2: Energy consumed by a robot doing the same operation in different times (Paryanto et al., 2015).

2.1.3 Pneumatic Systems and Actuators

Pneumatic actuators are devices in which energy of compressed air is used for car- rying out mechanical work like linear, rotary or oscillation movement (Funakubo, 1991).

Pneumatic systems are easy to maintain, clean, cheap, etc. thus, there are used in industrial automation systems. Energy efficiency improvements of pneumatic control systems can be made with different methods. Jihong, Wang, and Liau (2000) show in their investigation that velocity profiles of servo-pneumatic systems affect their energy efficiency. The results show that the system with a sinusoidal profile is the one that wastes less compressed air.

Another method of improving energy efficiency is calculating the exergy. Eret et al. (2012) made an approach of consumption without taking into account the dy- namic behaviour of the system and leakage air.

In a production line with a pneumatic actuator the quality of compressed air influences the quality of the final product. Besides, higher pressures increase leakage and this, increases the air consumption, so the cost. If in the production line is an actuator that needs higher pressure than others, the best solution for that is to install a smaller high-pressure unit or an amplifier (pneumatic booster) for that actuator instead of increasing the pressure of the whole factory compressed air system (Šešlija et al., 2012).

Practical studies made by Šešlija et al. (2016), Ignjatovi´c et al. (2013), and Šešlija et al. (2012) show that the best way to save energy in a complex robot cell with electric and pneumatic actuators is optimising first, parameters that influence the electricity consumption and after, parameters that affect the air consumption. In this way, the maximum optimisation of the cell is obtained.

2.2 Effects on Quality and Application Scenarios

There are many different application scenarios in industry and in each one the effects

on quality are different. In this thesis will be discussed the most used ones.

6 Chapter 2. Theoretical Framework and Literature Review

2.2.1 Material Handling

The most energy consuming component of material handling systems are the con- veyors. Zhang and Xia (2011) after building a model of a conveyor, they made an energy efficiency optimisation. Simulating six different optimisation problems that may have the conveyor system, they realise that the variable speed control can save energy. Further, the conveyor deviates from its design operation condition the more energy can be saved with the speed control system.

Masoudinejad et al. (2013) shows in their study that a permanent magnet syn- chronous motor for material handling applications (roller conveyors) needs a spe- cific rotational speed for each product with a particular weight to minimise the power consumption. This velocity depends on motor parameters. The method used enables a control system according to the weight of the product make the decision of the needed speed. This work does only take into account the steady state of the motor, i.e. there is no calculation of the energy consumption during the acceleration and deceleration.

Summarising, the power consumption of a conveyor system can be reduced con- trolling the speed of the motor. This control varies according to the weight of the products.

2.2.2 Machining

According to the work of Owodunni, Zhang, and Gao (2013), carried over models for aluminium milling, the energy used for machining a specific volume monotonously decreases with the increase in feed rate and spindle speed. The authors explain that energy consumption is non-conflicting with cost and time for any of the indepen- dent variables of the model. However, they show that it is conflicting with quality parameters and tool life. The overall conclusion is that milling parameters can be optimised to reduce cost, time and energy simultaneously, but slowing down a ma- chining process will not improve the efficiency of the machine but its quality result and mean time between reparations, at least for the conditions presented by the au- thors.

Empirical studies carried out by other researchers Mori et al. (2011) and Yan and Li (2013) show similar results, being the optimal solution for energy efficiency the operation with higher speed and feed rate values. Different types of carbon steel were the material for this experiments.

The hypothesis that power consumption can be reduced by slowing the produc- tion rate does not sustain in the area of metal machining. This can be achieved by slowing down the material handling processes between operations, but not the main tasks.

2.2.3 Assembling

Pellicciari et al. (2013) presented a method for energy consumption optimisation for robotic systems. Having manipulators electromechanical parameters and pre- scheduled trajectories, an optimal trajectory is determined. The results show that slowing down the speed reduce the energy consumption of the system, but slowing down as much as possible an operation, is not always good for reducing energy con- sumption. Is impossible to obtain very low speed due to it can affect negatively to the production rate.

Goya et al. (2012) propose an energy efficient method for SCARA (Selective Com-

pliance Assembly Robot Arm) robots. This method uses an adaptive elastic device

product weight to the robot repeatability and it was simulated in an assembly sta- tion. The results of the study show that when the robot is working at high speed the repeatability is worse. They realise that operating in higher speed levels could lead to decrease of productivity.

In summary, it can be say that in assembling, by slowing down the speed of the robot, energy consumption is reduced, but always between some limits, as it is said in the robots part.

2.2.4 Welding

Along with other parameters like welding current or arc voltage, speed affects the welding quality. Gery, Long, and Maropoulos (2005) realise that increasing the speed, the temperature of the welding plate decreases due to the heat is applied for a shorter period. This affects mainly to the fusion zone.

Ericsson and Sandström (2003) in their research arrived at the conclusion that mechanical and fatigue properties of friction stir welds are not affected by the com- mercial speed range. But, reducing the speed more than that, they obtained im- proved properties.

Kim et al. (2003) studied the relation between different process variables and bead penetration for a robotic CO

arc welding. Experimental results show that welding speed affects to bead penetration. When increasing the speed, more weld- ing current is needed to achieve bigger bead penetration. This means that reducing the velocity, less welding current is required to obtain greater penetration, so there is an energy saving when speed is reduced.

In the case of spot welding, for an optimisation of it, there is no need of changing the parameters of welding itself, but of the machine movements. As it is said in Section 2.1.2, this movements can be optimised.

2.2.5 Painting, Coating and Similar

Painting lines are the most power consuming stations of a automobile assembly fac- tory. Most of the energy in painting lines is spend during the drying process and for emissions treatment. Galitsky (2008) gives some ways to reduce energy con- sumption in a automobile assembly plant. One ways is to minimise he start up time. Painting stations need some start up time to obtain the temperature desired for the painting process. Minimising this start up time high amount of energy will be saved. An other way to reduce energy is to use an efficient ventilation system.

Using a computer-controlled ventilation demand control, ventilators will switch on only when they are needed and with the speed needed.

Analysing only the painting part, Li, Zhao, and Xie (2009) made an investigation of speed optimisation in a spray painting. Using genetic algorithm they optimise the painting speed and the results show that the speed optimisation improves the uniformity of the painting.

Painting works and similar are very delicate processes and more investigation is

needed to be able to adapt them to variable production rate.

8 Chapter 2. Theoretical Framework and Literature Review

2.3 Modelling and Simulation

One of the primary objectives of this project is to develop a way of optimising a system and apply it to a particular case of study. The behaviour of the system is in principle unknown and may follow a very complicated mathematic description. For the above, trying to find an analytical solution to the problem is discarded. The use of metaheuristic algorithms will allow the optimisation of systems without a deep understanding of their operation, but they can not be applied to running systems, so it is necessary to develop simulation models to perform the optimisation. Mod- els vary in scope and detail, and for this study, we will divide them into the three following categories.

2.3.1 Single Process Simulation

In this category will be grouped the most detailed and particular models. They are taken many parameters in consideration and are linked to the specifics of the oper- ation. Here come models that usually act with a set of fixed parameter given by a controller or are relatively independent. Since these models will normally be com- plex, diverse and computationally heavy,proper tools are needed to develop them.

In this work, Modelica is used.

Modelica

Modelica is a non-proprietary, object-oriented, equation-based language to model complex physical systems containing, e.g., mechanical, electrical, electronic, hydraulic, thermal, control, electric power or process-oriented subcomponents. It has complete and extensive libraries for lots of fields, and its basic usage consists on the connection of simple models.

To aid in the programming of the model, we will use the software OpenModelica as a graphical interface between Modelica and the user. This program also allows the use of optimisation tools with metaheuristics and multiobjective approaches.

2.3.2 Cell Simulation

What we are calling here a cell is a set of machines that interact with each other or whose actions are linked. Frequently, a single controller commands the cell respond- ing to the inputs of the system and external calls. Because of their nature, the models of these cells are much complex than the ones in the previous section, since they ac- count for more elements and control systems. These models can also be developed using Modelica, but there are much more suited programs like Simumatik3D.

Simumatik3D

Simumatik3D is a virtual reality simulator build to test PLC programs. It is capable of simulating complex cells with material handling elements, actuators and robots.

Although it is not designed for optimisation and does not take into consideration energy consumptions, thanks to the collaboration of its chief programmer new ca- pabilities will be added to solve these difficulties.

To implement energy calculations in Simumatik3D, the program will coordinate

models of every subsystem. The ones developed in this work will be exported and

controlled through the FMI standard.

of multiple cells. Once having simulated and optimised the cell, it can be reduced

to a much simpler statistical model of operation that can be later implemented in

programs like Arena or Plant Simulation to optimise buffer allocation and overall

performance.

Chapter 3

Methods

The optimisation of a whole production line is an extensive process that takes into account from the overall design to the smallest details. To be able to reduce the over- all production rate, every process has to be coordinated with the others to avoid fail- ures and defects. To ensure coordination, a hierarchical approach is proposed, where optimisation starts at every process and ends with the complete system. Commands start at the higher level and flow downwards to every machine.

Production administrator

Production line

Cell #1

Machine #1 Machine #2 · · · Cell #2 · · · Overall speed

of the line

Coordination among cells

Processes control

F

3.1: Tree diagram showing the flow of control commands

Optimiser

Independent processes

Objectives

Parameters

Process level

Optimiser

Dependent processes Production

cells

Parameters

Processes objectives

Processes parameters

Objectives

Cell level

Optimiser

Production line

Cells objectives

Cells parameters

Objectives

Line level

F

3.2: Global view of the optimisation process at every level

12 Chapter 3. Methods

This kind of approach is called a multilevel optimisation, and in this kind of problems, it has several advantages over a direct optimisation. In most cases, it will reduce the searching space for the optimisation algorithm. It allows an abstraction at higher levels, which allows for a better understanding for human operators. It also gives independence to the optimisation, avoiding the re-simulation of the whole system if some part is updated. For example, if one machine in a cell is changed for a different one only the modified cell and the line need to be re-optimised, and the line only if the Pareto front of the cell has changed significantly.

3.1 Process Optimisation

Every machine is responsible for making one or more operations to the product.

Each one of this processes is dependent on a set of parameters and environmental factors. The objective at this stage is to find the optimal set of parameters for a set of inputs that gives the minimum energy consumption at a given production rate.

The procedure to follow is to perform a multiobjective optimisation of the pro- cess to find the Pareto front between energy use and production rate

. The Pareto front is a set of solutions to a problem that can not be outperformed in one objec- tive without deteriorating other. In Figure 3.3 is shown an example of a Pareto front where it is intended to minimise the cycle time of an operation and its energy con- sumption for the parameters a, b and c.

a b

c

Cycle time

Ener gy consumption

F

3.3: Pareto front example. In the left a set of solutions in the parameter space, in the right their projection in the fitness space.

With the model of the machine, an optimisation for every operation is performed.

Many metaheuristic methods can be used (see Figure 3.4 for general classification), but population-based ones with global search strategy are preferred, such as evolu- tionary algorithms, particle swarm optimisation or ant colony approach among oth- ers. These algorithms can quickly (enhanced with parallel computation if needed) find the global minimums or maximums that conform the Pareto front.

Most processes in the industry are independent, i.e. they can be simulated and optimised by its own. In the event that a process is dependent or relays on other on- going operations, its optimisation should be delayed and be performed in conjunc- tion with the cell or line optimisation. This usually implies an increase in the number of parameters to optimise at these higher levels, but in some cases, the modelling of many processes as one dependent operation can reduce the number of variables.

1Other objectives can be added to the optimisation process, such a quality goals or similars. In this work, we will mainly focus on energy and cycle time

path in the parameter space is well-behaved.

Once the real machine is running with the optimised parameters, the overall cy- cle time or energy consumption may vary from the calculated in the simulations.

If the operation can support small parameter changes, a local search can be per- formed during operation with greedy algorithms like hill climbing, tabu search or simulated annealing. The data extracted from these experiments that are being per- formed while producing can be used to adjust the Pareto front and improve the general performance of the method.

The final goal of the individual process optimisation if to allow an abstraction at a higher level, where the number of parameters to control every step of the production will be highly reduced, normally to one.

Metaheuristics

Local search

Population

Trajectory

Naturally inspired

Dynamic objective function Evolutionary

algorithm

No memory

DirectExplicitImplicit

Tabu search

Particle swarm optimization

Simulated annealing Ant colony optimization

algorithms Evolution

strategy Genetic algorithm

Estimation of distribution algorithm Genetic

programming

Differential evolution

GRASP

Variable neighborhood search

Stochastic local search Iterated local search

Guided local search Scatter search

Evolutionary programming

F

3.4: Clasiffication diagram of most popular metaheuristics techniques (Johann nojhan Dréo, Caner Candan, Metaheuristics clas-

sification, CC BY-SA 3.0)

3.2 Cell Optimisation

As defined in section 2.3.2, a cell is a linked conjunction of machines operated by a

centralised controller. This controller is responsible for coordinating the machines,

taking care of the sequence or overlapping of processes, and setting the adequate

group of parameters, which can be stored and calculated in the controller itself or

consulted with an external call.

14 Chapter 3. Methods

When performing the optimisation of the cell, the parameters of the independent processes should not be in the search space. Instead, the algorithm will use the results of the process optimisation as variables for the cell optimisation (Figure 3.2).

For example, if the objectives of the process optimisation where energy consumption and cycle time, the cell optimiser will search through the cycle time, and get the process parameters from the already performed process optimisation.

The programming of every controller to adapt it to a variable production speed can be a tedious process that requires a full knowledge of the cell. Fortunately, most well-programmed systems are implemented to respond to a predefined set of vari- ables that can be redefined to be global variables dependent on an external server, or get a value from an auxiliary library or module. One important objective contem- plated in chapter 4 will be to minimise the modifications in the cell’s controller.

3.3 Line Optimisation

To ensure the correct flow of the products along the production line all the cells have to be coordinated. At the highest level, this coordination is carried out by a global controller, which have to indicate to each cell if it has to operate and at what speed.

This global driver must guarantee a correct functioning of the production pro- cess. It has the capabilities to achieve a uniform distribution of the products along the line and to reduce the total work in process. To achieve these objectives, the following capabilities are recommended.

• Slow the production to the bottleneck, especially before it. This would alleviate the use and need of buffers between operations.

• Increase the production speed of parallel machines when one of them is under maintenance, so the total production rate is maintained.

• If a process can not be slowed down to the required speed, the controller needs to activate it intermittently to match the overall production rate. To evaluate an appropriate frequency, the setting up cost must be considered.

• If the product position is actively monitored, when an unusual number of

products accumulates in a specific place the cells that come after can be ac-

celerated and the previous decelerated to alleviate the overload.

Chapter 4

Case Study

The method above described has been implemented in a cell of cylinder head pro- duction. It is made up mainly by three conveyor groups, which transport the prod- ucts and act as a buffer, a machining station, a checking station and an IRB 6620 robot. See Figure 4.1 for an overall snapshot of the cell.

Objects enter into the cell through the first conveyor group and they are trans- ported towards the robot, where they wait to be caught by it. Once the robot is free, it takes the object and leaves it in the machining station. The duration of this opera- tion depends on the kind of product that is at the station. Finished this process, the robot takes the object and leaves it on a second platform where a camera performs an automatic quality check. If no problem is detected, it continues the production process, but if the product has any failure, the robot leaves it in a separate conveyor group to remove the object from the production line.

An energy-velocity optimisation is pointless in the quality check station because it is a logic process. As seen in the machining section (2.2.2), slowing the working speed of a machining process does not yield benefits regarding energy consumption, so there will be no optimisation there either. The same elements conform all the conveyors in the transporting system so that, they will be modelled and optimised as a single process. The functioning of the robot can be divided into four independent processes, taking the units from the incoming conveyor to the drilling station, from there to the quality check and finally to the main path or the defective products conveyor. However, the robot here will be optimised as a single dependent process for illustrative purposes.

4.1 Conveyor Optimisation

The conveyors are the system that consumes the most energy in the cell, and they are always working at a higher production capacity than the robot, so an improvement in their performance will be very significant in the overall efficiency.

The model of the conveyor consists of two main parts, the driver motor and the conveyor load. All the conveyors are practically equal, thus only one model will be developed and optimised, using the non-propriety simulation language Modelica.

4.1.1 Conveyor Model

Figure 4.2 shows an overall diagram of the Modelica model for the conveyors. A

triphasic voltage source supplies the motor through a VFD, and it drives the bulk

load of the conveyor. The PLC is responsible for indicating the operating frequency

and starting acceleration, and the models return the total energy consumed.

16 Chapter 4. Case Study

F

4.1: Cell distribution

F

4.2: Conveyor model in Modelica

R

/s

−

u jX

F

4.3: Equivalent circuit of an induction machine

4.1.2 Motor Parametrisation

The motors used for the conveyors in this case study are ABB’s three phase squirrel cage motors, with four poles, 400 V and 1.1kW (more details in 3GBP092510-ASL).

Having the values obtained from different tests made by the manufacturer (see Test Report), and using the equivalent circuit of an inductive machine, the parameters of the motor are calculated.

The equivalent circuit of an inductive machine, Figure 4.3, shows the stator re- sistance (R

), stator inductance (X

), magnetising inductance (X

), rotor inductance referenced in the stator (X

) and rotor resistance also referenced in the stator (R

).

This last resistance is divided by the slip (s) which is calculated in equation (4.1) s = n

− n

n

(4.1) where n

is the synchronous speed of the magnetic field in revolutions per minute (rpm) and n

is the rotor speed in rpm.

From the equivalent circuit, Figure 4.3, the total equivalent resistance is defined and equation (4.2) is obtained

R

+ jX

+ jX

(jX

+

)

jX

+ jX

+

= u

√

3 I φ (4.2)

where u is the nominal voltage, I is the input current and cos φ is the power factor.

Three experiments are enough to fully determine the parameters of this circuit using the complex equation (4.2). The results of numerically solving this system are in Table 4.1.

Parameter Value R

4.22 Ω

X

15.7 mH

R

6.08 Ω

X

31.2 mH

X

426 mH

T

4.1: Obtained motor parameters

If there is no test report of the required motor, parameters can be calculated using

18 Chapter 4. Case Study

the data from the data sheet (Bruzzese et al., 2004). In an extreme case where there is not enough information about the machine, the best option will be to find similar motors and model them.

4.1.3 Load Parametrisation

In pursuit of applicability, the model of the conveyor load developed is very simple and can be determined entirely by the working data of the motors. The motivation for this simplicity is that in the industry is not common to have detailed informa- tion on the load of the motors, and taking the measurements for building the model should not stop the production. Considering the previous points, the conveyors have been modelled as an inertia and a linear dependent torque opposing the move- ment.

The torque parametrisation derives from the power consumption of the motor at standard load. Energy consumption is proportional to the torque, and from this relation, the simulation model and Newton-Raphson method can be used to get the stationary torque in a few iterations with (4.3).

P ≈ T ω → T

= T

− P

− T

ω

ω (4.3)

With a similar process, the inertia of the conveyor is extracted using the rise time of the motor and with the approximation that this rise time is proportional to the inertial.

4.1.4 Optimisation

The last significant value needed for the optimisation is the revolutions performed by the motor during a single process. Known the working length of the conveyor and the gear ratio between the motor rotation and linear speed, this value is easily obtainable. If the extension of the conveyor does not correlate with the usual work- ing distance, an average number of revolutions can be extracted from the monitored data of the motor.

The process is optimised using NSGA-II with a population of 50 individuals and, after a short number of generation, the solutions make up the Pareto front shown in Figure 4.4.

10 20 30 40 50

0 200 400

Frequency [Hz]

Acceleration [Hz/s]

5 10 15

4 6 8

Cycle time [s]

Specific ener gy [kJ]

F

4.4: Pareto front for the conveyor in the parameter and solu-

tion space

product through the length of the conveyor and cycle time the time needed to perform this operation. In the next level, cell optimisation, the optimiser will only search in the cycle time of this Pareto front to find the best fitting set of parameters for every throughput needed.

4.2 Cell Optimisation

Once all processes are optimised, the next step is to simulate and coordinate the cell. If there where only independent processes it would be a matter of coordination among them, and if there is any dependent process, as is the case, these processes have to be optimised with the cell. To simulate the cell, Simumatik3D is been used, capable of communicating with running PLCs, incorporate industrial robots and run advanced simulation models through Functional Mock-up Interface (FMI).

Conveyor model in Modelica

Robot in Robot- Studio

Conveyor

Robot Frequency

Total energy

Cycle time Proportional

speed Proportional

acceleration

Energy Velocity

Energy

Postition Digital IOs

Simumatik3D

F

4.5: Diagram of cell input and outputs

Conveyors in the cell are integrated with the FMI functionality. The robot is running in RobotStudio, and it provides an approximation of the energy consumed by the robot. The parameters in the robot optimisation are the proportional speed and acceleration of the joints referenced to the original program.

Simumatik3D simulates the whole cell with the parameters given by Modelica and RobotStudio. After finishing the simulation, it makes the sum of the consumed energy and the cycle time of a product, the time that a product spent in the cell is measured. This final values are sent to the user to have the final values of the cell.

4.2.1 Optimisation Algorithm

There are many optimisation algorithms that can be used in this cell. The one used

in this case is called Multi-Objective Stochastic Hill Climbing with Reset algorithm

(MOSHC-R) (Díaz and Suárez, 2001). It is based on the simple greedy algorithm Hill

20 Chapter 4. Case Study

Climber, but in the multi-objective context, it only considers a solution as better if it is a dominant solution. The hope with this algorithm is trying to travel over the Pareto front performing small changes in the parameters, giving a set of solutions with fewer evaluations than a population based algorithm. Of course, any other metaheuristic algorithm can be used at this level.

The cell optimiser is programmed in Python and connected to the PLC through an OPC server, similarly to what would be done in the industry.

4.2.2 Optimisation Results

After the optimisation, the produced Pareto front is shown in Figure 4.6. The cell is now able to adapt to the demand minimising the energy consumption while doing so. Is also remarkable that maintaining the original throughput of the cell the energy consumption has dropped by 49%, just by adapting the conveyors’ working pace to the production capacity of the robot.

28 30 32 34 36 38 40 42 44 46 48 50 60

70 80

Original configuration at (28,163)

Cycle time [s]

Specific ener gy [kJ]

F

4.6: Pareto front for the cell. The values from the pre-

optimised cell are out of the scope

21 Conveyor

model in Modelica

Pareto front NSGA-II

•

•

•

Process optimisation

Cell model in Simu- matik3D

Robot model in

Robot- Studio

Pareto front MOSHC-R

Cell optimisation

Cycle time Conveyor parameters

• • •

Line optimisation

F

4.7: Diagram of the complete optimisation at every level

Chapter 5

Conclusions and Discussion

5.1 Conclusions

From the literature review, is extracted that not every process is optimisable in the context of variable production speed. Mechatronic operations respond especially well to the optimisation, but machining operations only improve in terms of quality and tool life and do not get more energy efficient. The result of many other processes depend on the velocity of the operation, such in welding and painting, and these scenarios should be studied in every particular case.

The proposed method of optimisation at several levels reduce the search space for the optimisation if used sensibly, and allows the separation of elements for anal- ysis and command purposes. The most detailed and complicated simulations can be performed with specialised and diverse programs and methods, and a higher level is only a matter of coordination if there are no dependent processes.

The case study has been a useful way to put the concepts to the test and find potential issues. A simple model is preferred over a complex and detailed one since they are easier to determinate and use in the evolving industry and the results are usually good enough to perform an optimisation. In this particular case, the out- come of the optimisation has been very positive. Just by adapting the fastest pro- cesses to the bottleneck, i.e. the conveyors to the robot, the energy saving was of 49% over the measured processes, over 23 Wh per unit, which add up to hundreds of euros in savings over a year at the maximum production rate and only in one cell of the production line. If this cell turns out not to be the bottleneck in the production line the energy savings would be even greater, decreasing the consumption by 33%

compared to the optimised system at full speed or by 65% relative to the original cell.

5.2 Discussion and Future Work

As exposed before, the reduction of the working speed is not always a viable option, but it seems completely suitable for operations regarding motors and electric actu- ators, which conform a very considerable portion of the total energy consumed in the industry. Even if the core processes of the production can not be adapted to a variable production speed, the material handling systems can. There is a great po- tential for this idea in the automated industry since there will always be a bottleneck slowing down the production machines working faster than it, and this difference in production capacity is usually very significant.

This work only pretends to be an introduction to a new research branch. Meth-

ods and procedures should be revised and standardised. At a cell optimisation level,

24 Chapter 5. Conclusions and Discussion

in this work is decided to use Simumatik3D since the system was already imple- mented in it and have great capabilities to communicate with working PLCs and OPC servers. This is an intuitive and fast way of creating the model, and the PLC logic does not need to be re-implemented. However, if all processes in the cell are independent, i.e. the simulation outputs for every set of parameters have been cal- culated, a model of the cell through discrete event simulation can be achieved and be potentially much faster to optimise and the logic of the controller could be extracted from a flowchart or documentation.

The simulation results in this work are, as implied by their name, the product of a computer simulation and the method proposed here have not been implemented in a real system. All energy consumptions indicated in this work refer only to the working consumption of the modelled machines and stand-by consumption as well as other energy losses should be included in the optimisation algorithm. Even if these numbers do not come from a real cell, can be appreciated a significant improve- ment in the efficiency of the system and the process of optimising every process in a production line will surely lead to general improvements as a side effect.

A suggestion for next objectives in this line of investigation are the followings.

For this case study many tools are been used, Modelica and Simumatik3D for model design and simulations and Matlab and Python for optimisation and command pur- poses. Unifying all the needed tools under one platform with consistent use will facilitate utilisation and implementation of this work. For example, a Python library could be created using modules such as PyFMI and SimPy for model creation and simulation and DEAP for the optimisations. The implementation problems of this method were the most challenging aspect of it, so creating easy-to-use tools and fo- cusing on user development should be a priority for this idea to be implemented in the industry. Any optimisation or simulation was performed on a production line or factory level, so an adequate methodology should be developed for this purpose.

Also, more investigation needs to be done over non-mechatronic processes, as well as finding a viable application for this methodology with pneumatic and hydraulic actuators, if it is viable.

5.3 Sustainability Perspective

We would like to finish this thesis emphasising the sustainability potentials of this work, at environmental, economic and social levels. As indicated in the very title, the primary focus has been to reduce the energy consumption in industrial automated operations. If used extensively, the concept of reducing the working speed of ma- chinery will imply an immediate decrease in the energy consumed by the industry.

The environmental benefits are obvious; less consumption means less exhaustion

of natural sources and a reduction in the carbon footprint. Machinery working at a

lower rate also implies less wear, prolonging its lifespan and reducing the number of

reparations and pieces needed. These factors are not only good for the environment,

but for the owner of the machinery, who needs to spend fewer founds on energy

usage and maintenance. As any saving in the production process, this has repercus-

sions in society as more accessible products and services for the consumer. Good

results for the producer, consumer and the environment lead to a positive feedback

loop that promotes the implementation of this work and the benefit of all parties

involved.

Bibliography

ABB Group. RobotStudio. Version 6.0.

Abdelaziz, EA, R Saidur, and S Mekhilef (2011). “A review on energy saving strate- gies in industrial sector”. In: Renewable and sustainable energy reviews 15.1, pp. 150–

168.

Almeida, Aníbal T de, Fernando JTE Ferreira, and Dick Both (2005). “Technical and economical considerations in the application of variable-speed drives with elec- tric motor systems”. In: IEEE Transactions on Industry Applications 41.1, pp. 188–

199.

Ayani, Mikel. Simumatik3D. Version 0.9.5.

Beaty, H.W. and J.L. Kirtley (1998). Electric Motor Handbook. McGraw-Hill handbooks.

McGraw-Hill Education.

: 9780070359710.

Brossog, Matthias et al. (2014). “Energy consumption and dynamic behavior anal- ysis of a six-axis industrial robot in an assembly system”. In: Procedia CIRP 23, pp. 131–136.

Bruzzese, Claudio et al. (2004). “Numerical Identification of Parameters for Dynamic Analysis of Single-Cage Induction Motors Starting from Data-Sheet Quantities”.

In: IASTED Applied Simulation and Modelling Conference, ASM 2004, Proceedings of, pp. 28–30.

3GBP092510-ASL. Detailed information for: 3GBP092510-ASL.

: http : / / new . abb.com/products/ABB3GBP092510-ASL .

Díaz, Renio and Alejandro Rosete Suárez (2001). “A study of the capacity of the stochastic Hill Climbing to solve multi-objective problems”. In: Proceedings of the Third International Symposium on Adaptive Systems-Evolutionary Computation and Probabilistic Graphical Models, La Habana: Institute of Cybernetics, Mathematics and Physics. Citeseer, pp. 37–40.

Eret, P et al. (2012). “A practical approach to investigating energy consumption of industrial compressed air systems”. In: Proceedings of the Institution of Mechanical Engineers, Part A: Journal of Power and Energy 226.1, pp. 28–36.

Ericsson, Mats and Rolf Sandström (2003). “Influence of welding speed on the fa- tigue of friction stir welds, and comparison with MIG and TIG”. In: International Journal of Fatigue 25.12, pp. 1379–1387.

Funakubo, Hiroyasu, ed. (1991). Precision,Machinery and Robotics. Vol. 2. CRC Press.

Galitsky, Christina (2008). “Energy efficiency improvement and cost saving oppor- tunities for the vehicle assembly industry: an energy star guide for energy and plant managers”. In: Lawrence Berkeley National Laboratory.

Gery, D, H Long, and P Maropoulos (2005). “Effects of welding speed, energy input and heat source distribution on temperature variations in butt joint welding”. In:

Journal of materials processing technology 167.2, pp. 393–401.

Goya, Hidemasa et al. (2012). “Realization of high-energy efficient pick-and-place

tasks of scara robots by resonance”. In: Intelligent Robots and Systems (IROS), 2012

IEEE/RSJ International Conference on. IEEE, pp. 2730–2735.

26 BIBLIOGRAPHY

Greenyer, Joel et al. (2014). “Towards Synthesizing Energy-efficient Controllers for Modern Production Systems from Scenario-based Specifications”. In: Procedia Technology 15, pp. 388 –397.

: 2212-0173.

Ignjatovi´c, Ivana et al. (2013). “Optimisation of compressed air and electricity con- sumption in a complex robotic cell”. In: Robotics and Computer-Integrated Manu- facturing 29.4, pp. 70–76.

Jihong, W, J-Dai Wang, and Vui K Liau (2000). “Energy efficient optimal control of pneumatic actuator systems.” In: Systems Science 26.3, pp. 109–123.

Kim, IS et al. (2003). “A study on relationship between process variables and bead penetration for robotic CO 2 arc welding”. In: Journal of Materials Processing Tech- nology 136.1, pp. 139–145.

Li, Fa-zhong, De-an Zhao, and Gui-hua Xie (2009). “Trajectory optimization of spray painting robot based on adapted genetic algorithm”. In: Measuring Technology and Mechatronics Automation, 2009. ICMTMA’09. International Conference on. Vol. 2.

IEEE, pp. 907–910.

Test Report. M3BP 90LA 4 400VY 50Hz 1.1kW.

: https://library.e.abb.

com / public / 4af6d862bf024701a5b590c4cdcbadf0 / M3BP \ %2090LA \

%204\%20400VY\%2050Hz\%201.1kW.pdf .

Masoudinejad, Mojtaba et al. (2013). “Energy optimized speed regulation of perma- nent magnet synchronous motors for driving roller conveyors”. In: Power Elec- tronics and Drive Systems (PEDS), 2013 IEEE 10th International Conference on. IEEE, pp. 500–505.

Mecrow, B.C. and A.G. Jack (2008). “Efficiency trends in electric machines and drives”.

In: Energy Policy 36.12. Foresight Sustainable Energy Management and the Built Environment Project, pp. 4336 –4341.

: 0301-4215.

Modelica association. Functional Mock-up Interface (FMI). Version 2.0.

Modelica Association. Modelica.

Mori, M. et al. (2011). “A study on energy efficiency improvement for machine tools”.

In: {CIRP} Annals - Manufacturing Technology 60.1, pp. 145 –148.

: 0007-8506.

Offodile, O Felix and Kingsley Ugwu (1991). “Evaluating the effect of speed and payload on robot repeatability”. In: Robotics and computer-integrated manufacturing 8.1, pp. 27–33.

Open Source Modelica Consortium. OpenModelica. Version 1.11.0.

Owodunni, O.O., T. Zhang, and J. Gao (2013). “Energy consideration in machining operations - towards explanatory models for optimisation results”. In: 11th Global Conference on Sustainable Manufacturing : Proceedings. Innovative Solutions. Ed. by Günther Seliger. Berlin, Germany: Universitätsverlag der TU Berlin. Chap. 4.6, pp. 153–158.

Parr, Andrew (2011). Hydraulics and pneumatics: a technician’s and engineer’s guide. El- sevier.

Paryanto et al. (2015). “Reducing the energy consumption of industrial robots in manufacturing systems”. In: The International Journal of Advanced Manufacturing Technology 78.5, pp. 1315–1328.

: 1433-3015.

Pellicciari, M et al. (2013). “A method for reducing the energy consumption of pick- and-place industrial robots”. In: Mechatronics 23.3, pp. 326–334.

Riazi, Sarmad et al. (2015). “Energy optimization of multi-robot systems”. In: Au- tomation Science and Engineering (CASE), 2015 IEEE International Conference on.

IEEE, pp. 1345–1350.

Rockwell Automation. Arena.

Saidur, R. (2009). “A review on electrical motors energy use and energy savings”. In:

Renewable and Sustainable Energy Reviews 14.3, pp. 877 –898.

: 1364-0321.

Šešlija, Dragan et al. (2012). Increasing the energy efficiency in compressed air systems.

INTECH Open Access Publisher.

Šešlija, Dragan D et al. (2016). “Improving energy efficiency in compressed air systems- practical experiences”. In: Thermal Science 00, pp. 22–22.

Siemens PLM Software. Plant Simulation.

Motor energy saving. Ways to save on motor energy costs.

: http://web.archive.

org / web / 20081201114252 / http : / / www . aps . com / main / services / business/WaysToSave/BusWaystoSave_16.html (visited on 2008).

Yan, Jihong and Lin Li (2013). “Multi-objective optimization of milling parameters the trade-offs between energy, production rate and cutting quality”. In: Journal of Cleaner Production 52, pp. 462 –471.

: 0959-6526.

Zhang, Shirong and Xiaohua Xia (2011). “Modeling and energy efficiency optimiza-

tion of belt conveyors”. In: Applied Energy 88.9, pp. 3061–3071.