Application of Artificial Neural Networks in Condition Based Predictive Maintenance

Contemporary worldwide industrial environment with high demands of customers requires a robust, reliable and lean manufacturing processes.Maintenance is an activity required to keep the assets close to their original condition to make sure the asset can fulfil the function it was developed or even purchased for.There exist three strategies of maintenance: Corrective, Preventive and Predictive.

Corrective maintenance

Corrective maintenance, known as a run-to-failure is the one that is activated in case of machine failure . Such an activity is a must and leads to fast recovery of running Process.

Preventive maintenance

Preventive maintenance is also called condition based maintenance. It is based on the knowledge from machine manufacturers, experience of maintenance engineers and lessons learnt from past service incidents and means a set of activities to be done within a certain period on a set regular frequency (daily, weekly, monthly and annually).

Predictive maintenance

The predictive maintenance brings the new look and a philosophy on the maintenance strategies to achieve the maximum lifetime of machines while the risk of machine failure is minimised.

Designs and topologies Of ANN

ANNs have different topologies suitable for solving specific problems. It can be from a simple feed-forward network where the signal flows directly from inputs to outputs in one direction only till the recurrent or back-propagation networks where the signals beside the direct flow can make the loops to a previous layer or even within the same layer.

Work with artificial neural networks is always connected with processing of high vol-ume of data which requires advanced commercial and open-source software tools.

Applications of ANN

• Well trained ANN can cover noisy or incomplete data or is even suitable to predict unknown data based on defined parameters.
• Well trained ANN can cover noisy or incomplete data or is even suitable to predict unknown data based on defined parameters
• Industrial usage can be found in manifold intelligent sensory or vis-ual systems or in applications of process control.
• Applications of ANN in biomedicine are focused on early disease diagnosis in radi-ology, CT colonography or EEG.

ANN application in predictive maintenance

Artificial neural networks show promising results as a robust tool for evaluation of this data in order to support predictive maintenance activities.There exist a lot of papers focused on application of ANN in maintenance.

Mechanical damage and crack detection

Their method was capable of evaluating the risk of forming cracks using radial basis function neural networks (RBF) and suggested an implementation possibility of on-line monitoring.The data about the crankshaft rotation came from accelerom-eter, photo tachometer and the simulations were done with the new bearing and with bearings combinating artificial damage of inner race, outer race and a ball. For fault classification multi-layer perceptron with the conformance of 80% was successfully used.

Early detection of faulty electrical devices

This method profits from the phenomenon that every

object emits infrared radiation. In industrial maintenance, the application of infrared thermography is suitable for example for detection of heat losses in detection of heat of the electrical equipment, transformer load, etc.

This procedure advances from the fact, that it is robust against the differences of subject emissivity on measuring

its temperature. Normalisation of the images was done by conversion to grayscale bit-maps where focused regions were selected for detailed analysis.They achieved up to 83% of sensi-tivity rate on defect classification while parallel discriminant analysis prediction produced better results on specificity and accuracy.

Detection of faults on pneumatic systems

Pneumatic systems consisting of various pneumatic cylinders are used as cheap and reliable instruments for machine driven parts where there is no requirement for precise positioning.Some of the selected sensors used were vacuum analogue pressure sensor, material handling arm pressure sensor where the data were classified into two classes, signal below or over the 3 V. The type of ANN used was adaptive resonance theory (ART) with back propagation topology.

Monitoring of robotic manipulator

Robotic manipulators are used for precise positioning of manufactured parts during transfer to the following process step or they can be equipped as an automated welding machines.

Due to complicated design and implementation in fully au-tomated process, these have to be well maintained and monitored mainly for early risk of fault analysis.

Conclusions

Artificial neural networks show strong potential in industrial applications, especially in predictive maintenance tasks. Due to need of well-trained person for ANN computa-tions, necessity of special software and availability of sensors able to capture and sys-tematically store the collected data it has to be properly evaluated whether the use in real production will be meaningful and profitable.

The advantage of these methods is the potential of effective equipment failure prevention and the increase of the OEE per-formance results.

Beside the research of different ANN applications the authors focus on applications of ANN in the automotive industry chain. The future research will be focused on injection.

We would like to thank Dr. S. T. Patil for guiding us.

Author’s

1. Dr. S. T. Patil
2. Sahil Adhav
3. Aditya Sood
4. Ayush Prasad
5. Aryama Dubey