Predictive Maintenance using Machine Learning capabilities of MATLAB

For understanding and using the use of MATLAB's AI capabilities for machinery maintenance applications, a prior knowledge of maintenance engineering is a must. A practicing Maintenance Engineer / Consultant can easily correlate with the title of present article, i.e., application of Machine Learning tools contained in MATLAB for effective maintenance practices.

There are many built-in ML algorithms in MATLAB which could be used for various engineering practices, Maintenance Engineering being one among those hundreds of engineering areas in which these tools could be deployed.

Predictive maintenance algorithm in MATLAB helps in making effective maintenance decisions based on large sets of complex data. It is extremely difficult to decipher through huge volumes of such complex data arising from bearings, motors, engines, moving parts, etc,., that, manually, it is almost impossible to identify fault patterns, if there are any.

Routine maintenance practices like Breakdown Maintenance (abandoned almost half a century ago or more), Scheduled Maintenance, Reactive Maintenance etc., can prove to be quite expensive in today's highly competitive industrial and commercial world. Many a times the profit margins are so less which can prove to the difference between profit and loss due to unscheduled downtimes, machinery breakdowns, expensive parts replacements etc.

However, all such happenings could be avoided by careful implementation of ML algorithms into Manufacturing and Maintenance systems. There are proven tools and algorithms available in MATLAB and other platforms which can help an organization to study, analyze and implement them which will not only help in reducing and eliminates costly downtime but also improve overall efficiency of the plant in many aspects.

Rapid developments in the area of AI & ML have revolutionized every aspect of our daily lives, be it personal, official, industrial, logistic, or so many other fields. Maintenance Engineering is no exception for this worldwide revolution in application of Artificial Intelligence and Machine Learning for effective and productive maintenance practices. One such platform which has come a long way in improving maintenance capabilities is MATLAB by Mathworks.

MATLAB has many ML functions which can be analyzed and applied for routine maintenance practices thereby enhancing the maintenance capabilities. Some of the readily available built-in ML algorithms in MATLAB are : Diagnostic Feature Designer, Classification Learner, Signal Analyzer, Neural Network Clustering, Neural Network Pattern Recognition, etc.

There area readily available solved examples to understand these applications in MATLAB. By understanding and solving those examples (available on their homepage), one can gain experience in applying these tools for their industries.

Though the ML algorithms in MATLAB are relatively easy to deploy, understanding the outcomes of these toolboxes could be quite a challenge. Specialized training is a must for introducing ML tools in to existing maintenance application. Success or otherwise, depends on proper training of maintenance professionals on these tools is a must.

Another important aspect of introducing these tools is proper use of various sensors which are present in today's complex machinery. They generate humongous amount of data, which needs to carefully filtered and stored by servers. These Machine Signatures contain both good and faulty data, arising from good machine components and faulty components, which need to be analyzed carefully for identifying the machine faults.

A typical industrial pump data may have several thousand data sets. If one pump can generate so much of data over a day or weeks period, the amount of data generating from a huge machine or service equipment can be anybody's guess.

Well! Considering these challenges, and tools available to overcome these challenges, one can understand the importance of introducing Machine Learning tools and techniques for their Production and Maintenance practices.

The block diagram shown here gives a flowchart for introducing one such MATLAB TOOLBOX for effective Predictive Maintenance plan. The steps involved are:

1) Acquire the machinery data like : vibration, sound, temperature, pressure, etc.

2) Pre-process the data : to eliminate unwanted data (noise), arrange in proper sequence and store in proper folders for accessing later.

3) Identify key condition indicator: Identify the critical indicator of fault like severe vibration levels, intense sound, sudden increase in temperature or speed, increased levels of derbies in lubricants, etc.

4) Train the model: Ready models are already available in MATLAB. One needs to train these models as per their individual requirements. Training models are also readily available in MATLAB. This is where Machine Learning capabilities of MATLAB come in handy which can effectively and easily scan through the vast sets of machinery signature data and identify critical patterns, which is almost impossible manually.

5) Finally, develop the algorithm, critically test it and after successful testing it, deploy it in existing maintenance application.

For further understanding of these tools by MATLAB, one can visit their homepage, navigate to "Solved Examples" and study the algorithms. By modifying those to match their individual requirements, the organization can successfully deploy the ML TOOLBOX in their manufacturing system and derive the benefits like reduced waste, downtime, faulty components, etc., and improve overall uptime, efficiency and productivity.