LS-SVM based spectral clustering and regression for predicting maintenance of industrial machines

Rocco Langone, Carlos Alzate, Bart De Ketelaere, Jonas Vlasselaer, Wannes Meert, Johan A.K. Suykens
2015 Engineering applications of artificial intelligence  
Accurate prediction of forthcoming faults in modern industrial machines plays a key role in reducing production arrest, increasing the safety of plant operations, and optimizing manufacturing costs. The most effective condition monitoring techniques are based on the analysis of historical process data. In this paper we show how Least Squares Support Vector Machines (LS-SVMs) can be used effectively for early fault detection in an online fashion. Although LS-SVMs are existing artificial
more » ... nce methods, in this paper the novelty is represented by their successful application to a complex industrial use case, where other approaches are commonly used in practice. In particular, in the first part we present an unsupervised approach that uses Kernel Spectral Clustering (KSC) on the sensor data coming from a vertical form seal and fill (VFFS) machine, in order to distinguish between normal operating condition and abnormal situations. Basically, we describe how KSC is able to detect in advance the need of maintenance actions in the analysed machine, due the degradation of the sealing jaws. In the second part we illustrate a nonlinear auto-regressive model (NAR), thus a supervised learning technique, in the LS-SVM framework. We show that we succeed in modelling appropriately the degradation process affecting the machine, and we are capable to accurately predict the evolution of In industrial processes, the detection and analysis of faults ensure product 2 quality and operational safety [26]. Traditionally, three ways to deal with sensory 3 faults have been used [30],[31],[32]: corrective maintenance, preventive main-4 tenance and predictive maintenance. Corrective maintenance is performed only 5 when the machine fails, it is expensive and safety and environmental issues arise. 6 Preventive maintenance [16] is based on periodic replacement of components, 7 which are then utilized in a non-optimal way. Predictive maintenance [5] can 8 be performed in a manual or automatic fashion. In the first case machines are 9 manually checked with expensive monitoring techniques and the components are 10 replaced according to their real status. In the second case a machine's status is 11 automatically inspected and maintenance is planned accordingly. The continuous 12 monitoring of machine parts leads to reliable and accurate lifetime predictions, 13 and maintenance operations can be fully automated and implemented in a cost 14 effective way. 15 Nowadays, in many industries several process variables like temperature, pres-16 sure etc. can be measured. These measurements give an information on the current 17 status of a machine and can be used to predict the faults due to deterioration of 18 key components [23, 10]. As a consequence, an optimal maintenance strategy can 19 be planned. 20 Condition monitoring using sensor data has been performed for long time by 21 means of basic methods like exponentially weighted moving average, cumulative 22 sum, principal component analysis (PCA) [9], [6]. Only in the past few years 23 engineers in companies started to get convinced to use more advanced techniques 24 for fault detection, like for instance SVMs approaches in semiconductor manu-25 facturing [25, 15]. In this realm, with the aim of contributing to bridge the gap 26 between academia and industry, we propose to use LS-SVMs for predictive main-27 tenance. LS-SVMs [28, 27] are an artificial intelligence technique characterized 28 by an high quality generalization capability [29], the flexibility in the model de-29 sign and a clear procedure for model selection. Basically, for a given machine 30 we can construct a reliable model of the degradation process to be able to opti-31 65 The remainder of this paper is structured as follows: Section 2 summarizes 66 the KSC model and the related model selection scheme, i.e. BLF (Balanced Line 67 438 vised approach through kernel spectral clustering (KSC), and a supervised learn-439 ing method, namely nonlinear auto-regression (NAR). In the first case we used the 440 data collected by accelerometers positioned on the jaws of a Vertical Form Fill and 441 Seal (VFFS) machine, from which the degradation process of the machine con-442 ditions has been inferred. For the time-series analysis with the NAR model data 443 acquired by a thermal camera, which directly measures the dirt accumulation in 444 the jaws, are processed. We showed that LS-SVM can successfully assess and pre-445 dict mechanical conditions based on sensor data, thanks to their ability to model 446 the degradation process. Moreover LS-SVM achieved higher performance than 447 basic methods, which are commonly used in practice to predict the forthcoming 448 faults. Finally, we proposed two options to use LS-SVM to schedule maintenance 449 on the packing machine with different associated costs. 450
doi:10.1016/j.engappai.2014.09.008 fatcat:a74vgxaid5dt7hg2273qgndr4e