### Present Remotely

Send the link below via email or IM

CopyPresent to your audience

Start remote presentation- Invited audience members
**will follow you**as you navigate and present - People invited to a presentation
**do not need a Prezi account** - This link expires
**10 minutes**after you close the presentation - A maximum of
**30 users**can follow your presentation - Learn more about this feature in our knowledge base article

# A Real-Time Predictive Maintenance System for Machine Systems - An Alternative to Expensive Motion Sensing Technology

No description

by

Tweet## Joan Montraveta

on 14 December 2012#### Transcript of A Real-Time Predictive Maintenance System for Machine Systems - An Alternative to Expensive Motion Sensing Technology

THE AUTHORS OF THE ARTICLE ARE :

Dheeraj Bansal

David J.Evans

Barrie Jones School of Engineering and Applied Science

Aston University, Birmingham B4 7ET, UK JOAN MONTRAVETA MARGINET

RICARD MASANA GUIXÈ

GERARD PUIGORIOL ARNAUS THE ARTICLE WILL BE EXPLAINED BY : A REAL-TIME PREDICTIVE MAINTENANCE SYSTEM FOR MACHINE SYSTEMS

AN ALTERNATIVE TO EXPENSIVE MOTION SENSING TECHNOLOGY We can conclude that the use of neural network has been validated and the classification of the machine system parameters, on the basis of motion current signature is possible. As well as, it is a very useful technique in motor fault detection with a low failures percentage, concretely 2.4%. They designed a simulation model named "TuneLearn", which is capable of mapping the system parameters against the motion current feedback. The system allows to predict the drilling inertia in a surgeon micro-drilling procedure. The surgeons have full of control in drilling process, because the inertia changes and has been detected of the order of 1 • 10-6 g/mm2. It's a very accurate system. CONCLUSION Eventually, we would like to say that this concept (neural network to predict machine parameters) has been proved by classification only inertia as a system parameter.

Future work lies in identifying other parameters to be classified using the motion current signature, for instance friction or gravitation torque. The experiment consists in classify five different motor loads given

the motion current signature.

They started reading the motion signatures and firstly they collected 5000 signatures with a sample size of 900.

To reduce the dimensionality of the data and achieve the best prediction machine parameters, they used the following steps:

1. PRINCIPAL COMPONENT ANALYSIS (PCA)

2. PLOT_1 : EIGEN VALUES OF PCs vs NUMBER OF PCs

3. CHOOSE THE BEST CHOICE

4. MULTILAYER PERCEPTRON (MLP)

5. PLOT_2: MINIMUM VALIDATION ERROR vs NUMBER OF HIDDEN UNITS IN MLP

6. CHOOSE THE PARAMETERS OF NEURAL NETWORK AND ANALYZE THE TABLE_1

7. TABLE_2: CONFUSION MATRIX USING TEST DATA

8. WILL BE CORRECT THE CLASSIFICATION OF CONFUSING MATRIX ? PROOF OF CONCEPT 1.PRINCIPAL COMPONENT ANALYSIS This analysis is a mathematical procedure that uses an orthogonal transformation to convert a set of observations of possibly correlated variables into a set of values linearly uncorrelated variables called principal components (PCs).

Objectives:

1. To discover or to reduce the dimensionality of the data set.

2. To identify new meaningful underlying variables. 2. PLOT_1: EIGEN VALUES OF PCs vs NUMBER OF PCs The aim of this plot is to identify "K" principal components whose inclusion returns enough information.

On the other hand, in this plot they found three points (number of PCs) that have a noticeable step in the value of eigenvalues. 3. CHOOSE THE BEST CHOICE When they found the previous three points, they can calculate the cumulative contribution of information of each point.

If they use the 4 PCs point --> You obtain a 60% of information

If they use the 10 PCs point --> You obtain a 85% of information

If they use the 14 PCs point --> You obtain a 96% of information

It's obvious that they adopted the 14 PCs when they did the analysis.

Now, we can say that they managed to reduce the dimensionality of data from 900 to 14. 4. MULTILAYER PERCEPTRON (MLP) Is a feedforward artificial neural network model that maps sets of input data onto a set of appropriate output.

Consists of multiple layer of nodes in a directed graph, with each layer fully connected to the next one.

In this network, the information moves in only direction, forward, from the input nodes, through the hidden nodes, and the output nodes. 5. PLOT_2: MINIMUM VALIDATION ERROR vs NUMBER OF HIDDEN UNITS IN MLP They plotted a graph that represents the minimum validation error agains the number of hidden units in the MLP.

Now, if you pay attention in the plot the minimum error is when there are 15 hidden units. 6. CHOOSE THE PARAMETERS OF NEURAL NETWORK AND ANALYZE THE TABLE_1 When they finished the analysis of data, they chose the best options, these are the following ones:

--> 14 inputs --> 15 hidden units --> 5 outputs ( 1 of N coding ) Each output of the neural network correspond a class

which represented a load. 7. TABLE_2: CONFUSION MATRIX USING TEST DATA To evaluate the performance of the classifier, they used a "Confusion matrix" that shows the relation between true and predicted classes. 8. WILL BE CORRECT THE CLASSIFICATION OF CONFUSING MATRIX ? The neural network response to the test data, we can see that the confusion matrix shows 97'59% correct classification. THE ARTICLE AIMS

The main aim of this article is the creation of a low-cost technique for machine parameter estimation, that it motivates the development of a real-time predictive maintenance system.

This motivated an alternative approach for the development a System which generates the training data using simulation models. METHODS TUNE LEARN INTRODUCTION WHAT is sensing technology? Sensing technology consists in doing accurate control procedures, which can detect some failures and errors comparing the measures with data collection. PROBLEMS about the use of sensing technology: It is an expensive technology.

His maintenance is so complex, because it have a lot of components.

It is quite difficult to manage the big data collection.

His control is difficult and it needs an expert.

If somebody changes a sensor, the data collection will be unuseful. SOLUTION Design a real-time predictive method, which compares real measures with the operator measurements. IMPROVEMENTS Real-time method.

Faster than sensing technology.

More accurate.

Database isn't necessary.

Cheaper than sensing technology.

Less components and less complex. The component which never fails is the component which isn't in the machine. APLICATION This real-time method is used to estimate a real deviation about micro-drilling inertia.

This method has been applied in medicine, concretely in stapedotomy. It consists in doing holes in the ears (using micro-drilling) to insert very small components to improve the hearing about patients. ARTICLE HYPOTHESIS The main initial hypothesis that we did in this investigation was do a real-time predictive maintenance, which is better than doing a method based in data history.

Each ear is quite different and the database isn't useful in this procedure, the real-time method is a big progress.

In summary:

The Real-time predictive maintenance is better than database. The simulation model, named TuneLearn, was developed to generate the training data for the real-time predictive maintenance system and the neural network, for use in the real-time predictive maintenance system, was trained to predict the inertia deviations on the basis of the motor torque feedback.

The block diagram of the simulation model is shown in figure 1. The simulation model generates the motor current and velocity characteristics on the basis of system parameters, and is capable of modeling a fault condition which cannot be generated in an on-line environment.

As such, the model will be used to provide the mapping of system parameters and the current-time characteristic for all anticipated motor conditions. RESULTS The capability of the real-time predictive maintenance system is being used to predict the change in the drilling inertia on the basis of the motion current signature.

This system uses force and torque feedback from the tool point to estimate the state of the drilling process, allowing the system to accurately manage the procedure and execute a control strategy to reduce drill penetration. INDEX 1. INTRODUCTION 2. HYPOTHESIS 3. OBJECTIVES 4. METHODS AND RESULTS 5. CONCLUSION 6. REFERENCES A large amount of sensors used in

sensing technology STAPEDOTOMY TUNELEARN EXPLANATION REFERENCES The majority of the references belong to:

- IEEE ("Institute of Electrical and Electronics Engineers")

- IMCSD ("Incremental Motion Control Systems Society")

These both associations activities are dedicated to advancing technological innovation and excellence for the benefit of humanity. Thanks for your attention and now if you want, we will answer your questions. ?

Full transcriptDheeraj Bansal

David J.Evans

Barrie Jones School of Engineering and Applied Science

Aston University, Birmingham B4 7ET, UK JOAN MONTRAVETA MARGINET

RICARD MASANA GUIXÈ

GERARD PUIGORIOL ARNAUS THE ARTICLE WILL BE EXPLAINED BY : A REAL-TIME PREDICTIVE MAINTENANCE SYSTEM FOR MACHINE SYSTEMS

AN ALTERNATIVE TO EXPENSIVE MOTION SENSING TECHNOLOGY We can conclude that the use of neural network has been validated and the classification of the machine system parameters, on the basis of motion current signature is possible. As well as, it is a very useful technique in motor fault detection with a low failures percentage, concretely 2.4%. They designed a simulation model named "TuneLearn", which is capable of mapping the system parameters against the motion current feedback. The system allows to predict the drilling inertia in a surgeon micro-drilling procedure. The surgeons have full of control in drilling process, because the inertia changes and has been detected of the order of 1 • 10-6 g/mm2. It's a very accurate system. CONCLUSION Eventually, we would like to say that this concept (neural network to predict machine parameters) has been proved by classification only inertia as a system parameter.

Future work lies in identifying other parameters to be classified using the motion current signature, for instance friction or gravitation torque. The experiment consists in classify five different motor loads given

the motion current signature.

They started reading the motion signatures and firstly they collected 5000 signatures with a sample size of 900.

To reduce the dimensionality of the data and achieve the best prediction machine parameters, they used the following steps:

1. PRINCIPAL COMPONENT ANALYSIS (PCA)

2. PLOT_1 : EIGEN VALUES OF PCs vs NUMBER OF PCs

3. CHOOSE THE BEST CHOICE

4. MULTILAYER PERCEPTRON (MLP)

5. PLOT_2: MINIMUM VALIDATION ERROR vs NUMBER OF HIDDEN UNITS IN MLP

6. CHOOSE THE PARAMETERS OF NEURAL NETWORK AND ANALYZE THE TABLE_1

7. TABLE_2: CONFUSION MATRIX USING TEST DATA

8. WILL BE CORRECT THE CLASSIFICATION OF CONFUSING MATRIX ? PROOF OF CONCEPT 1.PRINCIPAL COMPONENT ANALYSIS This analysis is a mathematical procedure that uses an orthogonal transformation to convert a set of observations of possibly correlated variables into a set of values linearly uncorrelated variables called principal components (PCs).

Objectives:

1. To discover or to reduce the dimensionality of the data set.

2. To identify new meaningful underlying variables. 2. PLOT_1: EIGEN VALUES OF PCs vs NUMBER OF PCs The aim of this plot is to identify "K" principal components whose inclusion returns enough information.

On the other hand, in this plot they found three points (number of PCs) that have a noticeable step in the value of eigenvalues. 3. CHOOSE THE BEST CHOICE When they found the previous three points, they can calculate the cumulative contribution of information of each point.

If they use the 4 PCs point --> You obtain a 60% of information

If they use the 10 PCs point --> You obtain a 85% of information

If they use the 14 PCs point --> You obtain a 96% of information

It's obvious that they adopted the 14 PCs when they did the analysis.

Now, we can say that they managed to reduce the dimensionality of data from 900 to 14. 4. MULTILAYER PERCEPTRON (MLP) Is a feedforward artificial neural network model that maps sets of input data onto a set of appropriate output.

Consists of multiple layer of nodes in a directed graph, with each layer fully connected to the next one.

In this network, the information moves in only direction, forward, from the input nodes, through the hidden nodes, and the output nodes. 5. PLOT_2: MINIMUM VALIDATION ERROR vs NUMBER OF HIDDEN UNITS IN MLP They plotted a graph that represents the minimum validation error agains the number of hidden units in the MLP.

Now, if you pay attention in the plot the minimum error is when there are 15 hidden units. 6. CHOOSE THE PARAMETERS OF NEURAL NETWORK AND ANALYZE THE TABLE_1 When they finished the analysis of data, they chose the best options, these are the following ones:

--> 14 inputs --> 15 hidden units --> 5 outputs ( 1 of N coding ) Each output of the neural network correspond a class

which represented a load. 7. TABLE_2: CONFUSION MATRIX USING TEST DATA To evaluate the performance of the classifier, they used a "Confusion matrix" that shows the relation between true and predicted classes. 8. WILL BE CORRECT THE CLASSIFICATION OF CONFUSING MATRIX ? The neural network response to the test data, we can see that the confusion matrix shows 97'59% correct classification. THE ARTICLE AIMS

The main aim of this article is the creation of a low-cost technique for machine parameter estimation, that it motivates the development of a real-time predictive maintenance system.

This motivated an alternative approach for the development a System which generates the training data using simulation models. METHODS TUNE LEARN INTRODUCTION WHAT is sensing technology? Sensing technology consists in doing accurate control procedures, which can detect some failures and errors comparing the measures with data collection. PROBLEMS about the use of sensing technology: It is an expensive technology.

His maintenance is so complex, because it have a lot of components.

It is quite difficult to manage the big data collection.

His control is difficult and it needs an expert.

If somebody changes a sensor, the data collection will be unuseful. SOLUTION Design a real-time predictive method, which compares real measures with the operator measurements. IMPROVEMENTS Real-time method.

Faster than sensing technology.

More accurate.

Database isn't necessary.

Cheaper than sensing technology.

Less components and less complex. The component which never fails is the component which isn't in the machine. APLICATION This real-time method is used to estimate a real deviation about micro-drilling inertia.

This method has been applied in medicine, concretely in stapedotomy. It consists in doing holes in the ears (using micro-drilling) to insert very small components to improve the hearing about patients. ARTICLE HYPOTHESIS The main initial hypothesis that we did in this investigation was do a real-time predictive maintenance, which is better than doing a method based in data history.

Each ear is quite different and the database isn't useful in this procedure, the real-time method is a big progress.

In summary:

The Real-time predictive maintenance is better than database. The simulation model, named TuneLearn, was developed to generate the training data for the real-time predictive maintenance system and the neural network, for use in the real-time predictive maintenance system, was trained to predict the inertia deviations on the basis of the motor torque feedback.

The block diagram of the simulation model is shown in figure 1. The simulation model generates the motor current and velocity characteristics on the basis of system parameters, and is capable of modeling a fault condition which cannot be generated in an on-line environment.

As such, the model will be used to provide the mapping of system parameters and the current-time characteristic for all anticipated motor conditions. RESULTS The capability of the real-time predictive maintenance system is being used to predict the change in the drilling inertia on the basis of the motion current signature.

This system uses force and torque feedback from the tool point to estimate the state of the drilling process, allowing the system to accurately manage the procedure and execute a control strategy to reduce drill penetration. INDEX 1. INTRODUCTION 2. HYPOTHESIS 3. OBJECTIVES 4. METHODS AND RESULTS 5. CONCLUSION 6. REFERENCES A large amount of sensors used in

sensing technology STAPEDOTOMY TUNELEARN EXPLANATION REFERENCES The majority of the references belong to:

- IEEE ("Institute of Electrical and Electronics Engineers")

- IMCSD ("Incremental Motion Control Systems Society")

These both associations activities are dedicated to advancing technological innovation and excellence for the benefit of humanity. Thanks for your attention and now if you want, we will answer your questions. ?