The use of neural networks in statistical process control charts

Abstract

Neural networks (NNs) are massively parallel computing mechanism emulating a human brain. It has been proved that they had a satisfactory performance when they were used for a wide variety of applications. In the recent years, the efficiencies that provided the NNs also began to be applied in statistical process control (SPC). SPC charts have become one of the most commonly used tools for monitoring process stability and variability in today's manufacturing environment. These tools are used to determine whether the process is statistically under or out of control but in some cases such as the presence of autocorrelation as well as the presence of a specific pattern in the data do not provide the possibility of correctly and quickly detecting and classifying the existing fault. These problems have led many researchers to propose alternative methods for monitoring processes such as the use of NNs. In this paper, we discuss issues concerning the combination of both tools. Specifically, we study the NNs for the detection and determination of mean and/or variance shifts as well as in pattern recognition in the SPC charts. Furthermore, the use of NNs when the data are correlated is discussed. Finally, the use of NNs in multivariate control charts is addressed. The networks architectures that were used for each case, the way of operation and the performance of the proposed NNs applications are pointed out.

Methods for interpreting the out‐of‐control signal of multivariate control charts: A comparison study

“…Hwarng proposed an NN‐based method to detect out‐of‐control signals and identify which variable caused the out‐of‐control signal for the bivariate case. A survey on the applicability of NNs in SPM charts may be found in Psarakis …”

“…Although different ANN configurations are applied, the authors argue that for the detection of mean shifts, the following parameters should be considered: (1) the size of the input data; (2) the number of layers; (3) the number of hidden nodes; (4) the applied activation function; (5) the size of training file; (6) the length of the training, etc. The significance of these parameters in the interpretation of out‐of‐control signals is indicated in Psarakis …”

“…The significance of these parameters in the interpretation of out-of-control signals is indicated in Psarakis. 13 The basic methodology that is applied to detect out-of-control signals is comprised of the following steps:…”

Applying Two-Stage Neural Network Based Classifiers to the Identification of Mixture Control Chart Patterns for an SPC-EPC Process

“…Because autocorrelation widely exists in practical chemical or continuous processes [10,11,[25][26][27], the autocorrelation structure should be included in the process model. Therefore, the EPC is usually employed to compensate for effects of the autocorrelation and disturbances.…”

Economically Optimum Design of a Synthetic Chart

“…Recent papers based on statistical performance are, among others, Castagliola et al , De Magalhaes et al , Costa et al , Weiß & Atzmüller, Reynolds & Stoumbos, Schoonhoven & Does and Wu et l . Psarakis has studied the use of neural networks in statistical process control. However, Surtihadi & Raghavachari have stated that purely statistical model is not necessarily optimal from the cost (economic) point of view.…”