visual data mining of faults in machining process based on...

Proceedings of the 2016 International Conference on Industrial Engineering and Operations Management Kuala Lumpur, Malaysia, March 8-10, 2016

Visual data mining of faults in machining process based on machine learning

Jose luis Gonzalez rubio Department of Retail Banking Consulting,

Management Solutions, Torre Picasso 28020, Madrid, Spain

[email protected]

Yasser Shaban Department of Mechanical Design,

Faculty of Engineering, Helwan University, P.O. Box 11718, Cairo, Egypt

[email protected]

Soumaya Yacout Department of Mathematics and Industrial Engineering,

Polytechnique Montréal, Québec, H3C 3A7, Montréal, Canada

[email protected]

Abstract— Computing algorithms and technology are providing organizations and companies new methods in order to achieve their goals. Understanding complex physical phenomena, in which multiple variables are interacting, over time leads to advancement in many engineering fields. This understanding is based on processing huge amounts of readings and data. In this paper, we show the power of data visualization, when using many machining process’ sensors data in order to understand and to analyze the machining outcomes. Information is extracted from experimental results. Logical Analysis of Data (LAD) is used as knowledge extraction approach, which is presented in the form of characteristic patterns. The input data, the outcomes and the patterns are presented visually by using some visualization tools. We conclude with a discussion of the potential use of the data visualization.

Keywords— data mining; machining; logical analysis of data; data visualization.

I. INTRODUCTION

Data visualization is defined as the visual interpretation of the information obtained from data concerning complex relationships in multidimensional space. Graphic tools are used to illustrate the knowledge about some physical phenomena hidden in a dataset .Visualization aims at transforming knowledge from a format ready for computation into a format ready for human perception, cognition, and communication, thus presenting data to an observer in a way that yields insight and understanding. The field of information visualization, being related to many other diverse disciplines (for example, engineering, graphics, statistical modeling) suffers from not being based on a clear underlying theory. The absence of a framework for information visualization makes the knowledge that is extracted from a dataset, difficult to describe, validate and defend. Information visualization can be viewed as a communication channel from a dataset to the cognitive processing center of the human observer. This emphasizes the need to employ concepts from the theory of data communication as a mechanism for evaluating and improving the effectiveness of information visualization techniques.

Due to the importance of this subject, currently some researchers are undertaken in a new field called visual data mining, mainly in engineering and medical problems. For example, in [1], the authors used visual data mining to enhance the simple tools in statistical process control. The case study was used to show quality visualization toolkit to allow practitioners to implement some of these visualization tools without the need for training, extensive statistical background, and/or specialized statistical software. In [2], the core functional components, and a visualization module of data mining technology’s system are described. The paper provides a useful reference to visual data mining techniques. In [3], an overview of interactive visual data mining techniques for knowledge discovery is presented. The paper highlights the research on data visualization, visual

887© IEOM Society International


analytics techniques, and applications including a perspective on the field from the chemical process industry. In [4], Software agents paradigm in automated data mining for better visualization using intelligent agents are presented. In [5], the authors describe an ongoing effort to integrate information visualization techniques into the process of configuration management for software systems. In [6] the authors discuss a various types of 3D visualization and Visual Data Mining (VDM) techniques to analyze the time-dependent change of magnetic field line's topology more effectively and efficiently. In [7], Visual data mining for identification of patterns and outliers in weather stations' data are presented. The authors present and discuss the techniques' details, variants, results and applicability on this type of problem. Traffic predictions is presented in[8], and approaches of visual data mining and artificial intelligence are introduced to solve traffic sensitive circumstances and congestion risk.

In this paper, the objective is to show the power of data visualization in conveying knowledge about faults in machining process. The input data, the output knowledge about the regions where faults take place and about the patterns corresponding to these faults, are all presented visually. The information that is extracted from the experimental results is presented in the form of characteristic patterns which are used are shown visually. In section 2, the case study and the experimental procedure which was carried out in order to collect data that is used for visualization, is presented. In section 3, visualization of different datasets that represent the input data, the outcomes, and the extracted knowledge in the form of patterns are presented. Concluding remarks are given in section 4.

II. CASE STUDY

A. OverviewIn this paper the data is extracted from a machining process of Carbon fiber reinforced polymer (CFRP). CFRP is an

important composite material which has special properties; its unique properties make it the backbone of some industries such as aerospace, sporting, automotive and aircraft structure [9]. The non-homogeneous composition and abrasive properties of reinforcing fibers yield some difficulty the machining process. The measured machining conditions are used to evaluate the quality and the geometric profile of the machined part. Here, the measured machining conditions are presented in terms of machining forces. A pattern-based machine learning approach is used to detect the characteristic patterns that differentiate between normal state and faulty state of the machining process. The approach is called Logical Analysis of Data (LAD). LAD finds the characteristic patterns which lead to conforming products and those that lead to nonconforming products. LAD is used for online control of a simulated routing process of CFRP [10], and used to diagnosis of drilling outcomes of CFRP [11]

LAD is a knowledge extraction approach that allows the classification of phenomena based on knowledge discovery and pattern recognition. It is applied in two consecutive phases, training phase, where part of the database is used to extract patterns of some phenomenon, and the testing phase, where the rest of the database is used to test the accuracy of previously trained knowledge. LAD uses a supervised learning technique; this means that the historical data or the database contains the variables and their corresponding outcomes (classes). LAD was used for the first time in fault diagnosis in 2007 [12], and continued to be used successfully for fault diagnosis [10, 13].The main steps of the LAD are the binarization of data, the pattern generation,and the theory formation.

B. Experimental description and data

The composition of the tested CFRP composite is quasi isotropic laminate comprising 35 plies of 8-arness satin wovengraphite epoxy prepreg with a final cured thickness of 6.35 0.02 mm [10].

Two tests were performed as follows:

1. The routing tests are performed using:• Four values of spindle speed (rpm): 10,000, 20,000, 30,000, and 40,000.• Three values of feed (mm/min): 250, 500, and 1,000.• Three values of tool overhang lengths (LT): LT1 = 38 mm, LT2 = 31 mm, and LT3 = 24 mm.• The experiments are repeated each 32mm of cutting distance for three times. As such, we have three values of cutting

distance (LC): 32, 64, and 96 mm.

In total, we have three feed rates ( ), four cutting speed ( ), three overhang length (LT), and three cutting distances (LC); therefore, the total number of observations (experiments) is 108, sample of the experimental results are showed in Fig.1. A full factorial design of experiments is used. The cutting forces are measured using a Kistler dynamometer, and the temperatures are measured using Thermo Vision A20M infra-red camera. The conforming specifications of these qualities are as follows:

• Exit and entry delamination ≤ 1%.• Slot surface roughness right and left ≤ 1.2 μm.



2. The drilling tests are performed at five different speeds ( ) with seven different feeds ( ), which results in 35

observations as summarized in Fig. 2. In each observation, the thrust force (Fz), the cutting force (Fc) and the torque (M) are monitored and measured [11]. The conforming specifications of these qualities are as follows:

• Exit and entry delamination = 1.0 • Hole circularity at exit and at entry ≤ 0.2% • Hole diameter error at exit and at entry ≤ 0.02% • Hole surface roughness ≤ 0.5 µm.

Table I presents the output patterns, that divide the machining conditions and the sensors readings into two regions of conforming (positive) and nonconforming (negative) outcomes, are obtained by LAD for four models (A-1), (A-2), (B-1) and (B-2) in routing process. The characteristic patterns are found by the cbmLAD software[14]. Model (A-1) has controllable variables as inputs, namely cutting speed ( ), feed ( ), overhang length, and cutting distance. The output is the delamination which can be conforming or non-conforming to specifications. Model (A-2) has the same controllable variables as Model (A-1), as inputs, and the surface roughness, which can be conforming or non-conforming, as output. Model (B-1) has the monitored uncontrollable variables, namely the forces in three coordinates and the mean temperature Tmean, as inputs. The output is the delamination which can be conforming or nonconforming. Model (B-2) has the same monitored uncontrollable variables as Model (B-1), as inputs, but the output is the surface roughness which can be conforming or non-conforming.

TABLE I THE PATTERNS OBTAINED BY LAD [10]

The generated positive patterns illustrate the threshold boundaries for the controllable conditions that will always lead to conforming (positive) parts. The negative patterns that are formed with the uncontrollable variables are used to give an alarm indicating that the machining process is beginning to produce unacceptable products.

III. VISUALIZATION OF THE DATA As it is seen in the description of the case study, this situation represents a challenge in the visualization issue because of

the number of experiments, the controllable, and the uncontrollable variables. In the following sections we give the results of some visualization of the input data, the obtained outcomes and the patterns. In order to visualize the experimental data, Fig. 1 and Fig. 2 are presented. In order to extract meaningful information, the dataset is cleaned to clarify its comprehensiveness. Bar charts are added in the cells of the uncontrollable variables and outcomes in order to visualize the evolution of these values of uncontrollable variables with the evolution of the controllable variables. The software that is used is Microsoft Excel.

A binary table with a logic function is added on the right in Fig. 1 in order to show the outcomes’ compliance to the specifications. The red color (0) indicates the noncompliance, and the green (1) indicates the compliance. One of the important information that is observed in Fig. 1 is the non-linearity of all the relations between the controllable and the uncontrollable variables, and also between the inputs and the outcomes.



Fig. 1. Datasets of input data, the uncontrollable forces, the outcomes, and the compliance to specifications for routing.

In Fig. 2 all the observations are placed beside the values of the forces and torque. A bars chart has been placed in each of

the cells to show the change in the values of the uncontrollable variables with the change in the values of the controllable variables. It is interesting to look at the relation between the outcomes as functions of the forces (the uncontrollable variables). To achieve this purpose, a table with the thirty five tests’ observations, the values of the forces and the outcomes, is presented.

Fig. 2. Controllable, uncontrollable and outcomes values for drilling.

In order to show the successful compliance to specifications for the seven qualities, a surface plot is used in Fig.3. Each of the specifications is expressed by indicating the percentage of fails versus successes in the thirty five observations resulting



from the experimental tests. In the heptagon where each vertex is one specification, the uncolored area represents the percentage of success whereas the red one the fails.

Fig. 3. Percentage of fails (noncompliances) vs successes (compliances) in drilling process.

Fig. 4 represents a 3D surface plot using the number of successes or fails as the dependent variable. Employment of a pivot table is needed to tackle this figure, as it allows the counting of fails or successes among all the observations for each pair of spindle speed and feed. A color scale has been used as a qualitative attribute, utilizing green for the best situation (greater number of compliances) and red for the worst (non-compliances).

Fig.4. 3D Surface plot for drilling process.

Up to this point the outcomes, the compliance or noncompliance to the specifications have been set as a binary variable

where only two possibilities are shown, fail (0) or success (1). Since these outcomes have specific values, it is possible to establish a certain grade of closeness between the value obtained and the specifications aimed. In Fig. 5, the original table, with the values of the hole diameter error in % at Exit, have been substituted by the absolute value of the difference between the limit value of this specification (in this case 0.02) and the actual value obtained. Again a color scale is used to show the region where the specification value is respected and the regions where it is not and by how far. Darkness means proximity whereas



lightness means distance. At the same time the combinations of inputs that accomplish specifications are surrounded by a green line.

Fig.5 Proximity to constraint in entry Delamination

Similar work has been done for the other specifications; working in a parallel way as done in Fig.6, a percentage among the

seven tables can be calculated; representing in one table all the specifications at the same time.

Fig.6 Proximity to targets (all the specifications considered).

A. Relation with the specifications Figures 7 and 8 were created with the tool “pivot tables” of Excel. The two figures show the different outcomes as

functions of the controllable variables. They show the outcomes’ averages. At 3000 rpm, the acute change in average delamination is occurred for all values of feed.

Fig. 7 Average Delamination for length of cut 30mm in routing process.

0.005

0.010

0.015

0.020

0.025

0.030

0.035

0.040

0.045

0.050

10000 20000 30000 40000

Ave

rage

Del

amin

atio

n

Cutting Speed (rpm)

Length of cut 30

Feed 250. Average Exit DelaminationFeed 250. Average Entry DelaminationFeed 500. Average Exit Delamination"Feed 500. Aveage Entry Delamination"Feed 1000. Average Exit DelaminationFeed 1000. Average Entry Delamination



Fig. 8 Average surface Roughness for length of tool 38mm.

Another visualization has been done with the uncontrollable variables. The 108 outcomes values are represented as functions of the forces and the Tmean take. On Fig. 9 and 10 a plot for the Exit and Entry Delamination and Slot surface roughness respectively as a function of Fx is done. At 3000 rpm, the acute change in average surface roughness is occurred for all values of feed.

Fig. 9. Exit and Entry Delamination as a function of Fx.

Fig. 10 Surface Roughness right and left as a function of Fx.

2.5

3.5

4.5

5.5

6.5

7.5

8.5

10000 20000 30000 40000

Ave

rage

Sur

face

Rou

ghne

ss

Cutting Speed (rpm)

Length of tool 38

Feed 250. Average Roughness Right

Feed 250. Average Roughness Left







B. Patterns’ visualization With the software cbmLAD and the models (A-1), (A-2), (B-1) and (B-2), the patterns that distinguish the compliance

region from the noncompliance regions are shown in Fig. 11 and 12.

Fig 11. Positive Patterns model (A-1).

Fig 12. Negative Patterns model (B-1).

The aim of this work is to give the user an easy understandable visualization of the physical phenomena that affect the machining processes in terms of the controllable and controllable variables. With this purpose the patterns generated from the drilling process has been transformed into a surface plot where the regions that correspond to acceptable outcomes are easy to find. To illustrate, in Fig. 13, the hole diameter error in percentage at exit has been chosen. Table II gives the two negative patterns the four positive, each region delimited by each pattern is designed, and then put together in Fig. 13. Finally all the regions covered by the positive and the negative patterns are put together in the same chart of Fig. 14, so it is clear which are the regions where the combination of speed and feed produce an acceptable quality.



TABLE II THE PATTERNS OBTAINED BY LAD HOLE DIAMETER ERROR AT EXIT FOR DRILLING PROCESS Negative patterns for hole diameter error at exit Positive patterns for hole diameter error at exit Pattern 1: speed >10250 feed >80, <150 Pattern 2: Speed <6750 feed >40, <80

Pattern 1:speed feed >150 Pattern 2: speed >6750 feed <80 Pattern 3: speed <10250 feed >80 Pattern 4: speed feed <40

Fig.13. Hole diameter error in percentage at exit.



Once the regions of the positive and negative patterns are placed together, the result is:

Fig.14. Positive and negative patterns for hole diameter error in percentage at exit in drilling process.

In Fig. 15, the area in green shows the region covered by the two negative patterns which indicate the combination between the spindle speed and the feed that produces an acceptable outcome. In terms of compliance to the specification ‘exit delamination’, the four green points in the chart of Fig. 14 are the four green cells in the Fig. 15, all leading to acceptable quality.

Fig. 15. Accomplishment of the Hole Diameter error in % ar Exit.

C. Box Whiskers The set of data used in this paper give a good indication of the advantage of using LAD, which is a non-statistical, non-

approximate technique. LAD does not assume that the data belongs to any specific statistical distribution [16]. As box whiskers shows the relation between the outcome and each of the variables, one by one. Box plot is used to depict groups of numerical data through their quartiles graphically and display variation in samples of a statistical population without making any assumptions of the underlying statistical distribution. The creation of Box whiskers is obtained by using Matlab.



Fig. 16. Box whiskers for entry delaminarion.

In Fig. 16, four box whiskers are built in to show the relation between the Entry Delamination and each of the controllable variables.

IV. CONCLUSIONS Visualization is an essential aspect of decision making. It helps in representing the knowledge that is extracted from

datasets in a clear and comprehension way. The main objective of data visualization, according to common definitions is to “amplify cognition” by externalizing thought processes. As a tool for “crystallizing new knowledge,” visualization allows us to perceive and recognize patterns in data. This paper presented some examples of visualization in the field of machining. It presented, the input data, the outcomes with respects to some specifications, and the knowledge discovered in the form of patterns.

Information is extracted from experimental results, and is presented in the form of characteristic patterns which are used

in data visualization. Logical Analysis of Data (LAD) is used as knowledge extraction approach. The next research challenge is to visualize big data. The field of visualization is becoming more and more important because managing huge amounts of data is growing quickly, however the development of new visualization tools is still to be done.



REFERENCES

[1] H. D. Smith, F. M. Megahed, L. A. Jones‐Farmer, and M. Clark, "Using Visual Data Mining to Enhance the Simple Tools in Statistical Process Control: A Case Study," Quality and Reliability Engineering International, vol. 30, pp. 905-917, 2014. [2] G. Feng, Z. Li, and L. Zhang, "Researches on the Prototype Implementation of Visual Data Mining Techniques," International Journal of Database Theory and Application, vol. 7, pp. 131-138, 2014. [3] F. Stahl, B. Gabrys, M. M. Gaber, and M. Berendsen, "An overview of interactive visual data mining techniques for knowledge discovery," Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, vol. 3, pp. 239-256, 2013. [4] R. Jayabrabu, V. Saravanan, and K. Vivekanandan, "Software agents paradigm in automated data mining for better visualization using intelligent agents," Journal of Theoretical and Applied Information Technology, vol. 24, pp. 167-177, 2012. [5] L. Voinea and A. Telea, "Visual data mining and analysis of software repositories," Computers & Graphics, vol. 31, pp. 410-428, 2007. [6] D. Matsuoka and M. Ken, "3 D Visualization and Visual Data Mining," Journal of the National Institute of Information and Communications Technology, vol. 56, pp. 507-517, 2009. [7] J. R. M. Garcia, A. M. V. Monteiro, and R. D. Santos, "Visual data mining for identification of patterns and outliers in weather stations’ data," in Intelligent Data Engineering and Automated Learning-IDEAL 2012, ed: Springer, 2012, pp. 245-252. [8] W. Schneider and W. Toplak, "Traffic predictions with visual data mining and artificial intelligence," e & i Elektrotechnik und Informationstechnik, vol. 125, p. 232, 2008. [9] M. Rahman, S. Ramakrishna, J. Prakash, and D. Tan, "Machinability study of carbon fiber reinforced composite," Journal of Materials Processing Technology, vol. 89, pp. 292-297, 1999. [10] Y. Shaban, M. Meshreki, S. Yacout, M. Balazinski, and H. Attia, "Process control based on pattern recognition for routing carbon fiber reinforced polymer," Journal of Intelligent Manufacturing, 2014. [11] S. Yacout, M. Meshreki, and H. Attia, "Monitoring and control of machining process by data mining and pattern recognition," in 2012 Sixth International Conference on Complex, Intelligent, and Software Intensive Systems (CISIS),, 2012, pp. 106-13. [12] D. Salamanca and S. Yacout, "Condition based maintenance with logical analysis of data," 7e Congrès International de génie industriel, 2007. [13] Y. Shaban, S. Yacout, and M. Balazinski, "Tool wear monitoring and alarm system based on pattern recognition with Logical Analysis of Data," Journal of manufacturing science and engineering, 2015. [14] S. Yacout, D. Salamanca, and M.-A. Mortada, "Patent Cooperation Treaty PCT/CA2011/000876, No. Wo 2012/009804 A1," 2012.

BIOGRAPHY

Jose-luis Gonzalez-rubio is a consultant with the department of Retail Banking at a business consulting firm in Spain. He is currently studying M.Sc. in Business Consulting at Universidad de Comillas, Madrid, Spain. He holds a B.Sc, and M.Sc. degree in Industrial Engineering (level 7 in the European Qualifications Framework) from Universidad de Sevilla, Spain, and completed his training at École Polytechnique de Montréal, Canada. Yasser Shaban is Assistant Professor in the Department of Mechanical design Engineering at Helwan University in Cairo, Egypt. He holds a Ph.D. in industrial engineering from Polytechnique Montréal in Canada. He holds a B.Sc., and M.Sc. degree from Helwan University, Cairo, Egypt, in Mechanical Engineering. His research field is diagnosis of machining conditions based on artificial intelligence. He is a member of the Institute of Industrial Engineers.

Soumaya Yacout is a Professor in the Department of Mathematics and Industrial Engineering at Polytechnique Montréal in Canada. She holds a D.Sc. in operations research from Georges Washington University in U.S.A.; and a B.Sc., and a M.Sc. in Industrial Engineering from Cairo University in Egypt. Her research interests include Condition Based Maintenance, and distributed decision making for product quality. She is a senior member of the American Society for Quality; and a member of the Institute of Industrial Engineering, and the Canadian Operational Research Society.


visual data mining of faults in machining process based on...

Documents