برآورد باناسال جونز: یک الگوریتم معکوس برای سیستم نگهداری پیشگویانه زمان واقعی

An algorithm, to estimate the machine system parameters from the motion current signature, based upon non-linear time series techniques for use in the real-time predictive maintenance system is presented in this paper. Earlier work has introduced the use of a neural-network approach to learn non-linear mapping functions for condition monitoring systems. However, the performance of the neural-network largely depends upon the quality of the training data, and that of the quality and type of the pre-processing of the input data. A reverse algorithm called BJEST (Bansal–Jones Estimation), for estimating the machine input parameters using the motion current signature, has been designed and proven to be successful in estimating the macro-dynamics of the motion current signature. This motivated the enhancement of the predictive analysis to incorporate non-linear characteristic of the motion current signature. The results show considerable improvement in the estimation of the parameters using the enhanced BJEST algorithm due to estimation consistency, hence, improving the real-time predictive maintenance system.

Growing complexity of industrial manufacturing and the need for higher efficiency, greater flexibility, better product quality and lower cost have changed the face of the manufacturing practice. The need for efficiency and maximum production time creates a requirement for high reliability supported by an effective maintenance system. Functions such as maintenance that are seen as being ‘non-value adding’ are being continuously required to reduce costs, whilst keeping equipment running in an optimum condition. Effective monitoring can support cost reduction and efficiency improvement strategies [1]. Modern machines typically use some form of direct current (DC) motor for motion dynamics and the process described is based upon such motors. Earlier work has introduced an effective, real-time, predictive maintenance system based on the motion current signature using a neural-network approach [2] and [3]. The aim of the system is to localize and detect abnormal electrical conditions in order to predict mechanical abnormalities. One of the fundamental requirements for the successful application of a neural-network is the availability of relevant, information-rich training data. While an ideal solution would be to utilize training data collected from a real production system, it is impractical to scan the entire range of machine operations [2]. Thus, the use of a simulation model for generating the training data, covering harder to replicate machine conditions, like current limit over-run, was motivated [2]. A simulation model, TuneLearn,1 of a closed-loop form based on a PID controller was developed and shown to be capable of mapping the motion current signature to the system parameters [3]. Whilst validating the simulation model, a reverse algorithm called BJEST (Bansal–Jones Estimation), for estimating the machine input parameters using the motion current signature, was designed and proven to be successful in estimating the macro-dynamics of the motion current signature [3]. Also, the performance of the neural-network largely depends upon a number of factors including: (1) the quality of the training data; (2) quality and type of the pre-processing of the input data; (3) the type of the neural-network technique adopted; (4) the training methodology used. The success in estimating the macro-dynamics of the motion current signature using BJEST, referred to as the reverse algorithm here onwards, and the dependence of the performance of a neural-network approach upon training data, motivated the enhancement of the reverse algorithm to incorporate non-linear characteristic. This paper details the enhancements performed to the reverse algorithm to incorporate the non-linear characteristics of the motion current signature. However, the first step is to statistically test the motion current signature for non-linearity, which has been performed using surrogate data testing. The remainder of this paper is organized as follows: Section 2 discusses the surrogate data test as a means for testing for non-linearity in the motion current signature; Section 3 provides the results of the surrogate data testing; Section 4 describes the enhanced version of the reverse algorithm, which integrates non-linear noise reduction techniques to improve estimation performance; Section 5 provides the comparison of the performance of the new enhanced reverse algorithm against the linear reverser algorithm described in [3]; and Section 6 gives the conclusions.

Nonlinearity of the motion current signature has been established by using surrogate data test. It has been shown that the motion current signature obtained from a production machine does not have the same distribution as a linear stochastic process, thus confirming the presence of non-linear component in the signature. The reverse algorithm has been modified by including Schreiber noise reduction technique as a pre-processing technique to filter out the noisy non-linear component of the signature. It was established that the filtered signature has the same distribution as a linear stochastic process and, hence, the filtered signature can be used for further predictive analysis using linear reverse algorithm. Furthermore, it was shown that the prediction using the enhanced reverse algorithm is more consistent than before. Consistency of system parameter prediction coupled with improved understanding of the motion dynamics motivates the comparison of the performance of enhanced reverse algorithm against that of neural-network approach. Ongoing work is now concerned with the implementation of the reverse algorithm in a production environment to test its performance in real-time.