Research on electronic device identification based on small data sets

Abstract

The paper investigates the method of identifying the parameters of electronic components when selecting them for critical applications equipment based on failure predictors. This may be necessary when equipping important complex electronic systems with a high-quality component base. To ensure this, in the production of electronic components, it is often necessary to select more reliable, higher-quality components for such systems. One of the important approaches in this case is the choice of data processing methods and identification techniques. A significant part of such approaches requires the improvement of data processing methods, including clustering. The implementation of the identification technique is based on determining the number of clusters and clustering using wavelet transform. To solve the problem of approximating the current-voltage characteristics and choosing the identification vector of transistor characteristics, a radial basis neural network was used, and a multilayer perception was selected for further classification. After determining the number of clusters and grouping the output data using clustering based on the wavelet transform, the number of neurons in the neural network was reduced by almost one and a half times, and the classification time was reduced by more than thirty percent. These results allow us to recommend the developed identification method for diagnosing using “predictors” of failures for use in a wide range of practically important tasks in the control and diagnosis of electronic equipment and electronic engineering products in the case of variable parameters of the objects of diagnosis, with a high level of noise in the measurement data and with small volumes of the studied samples. The use of this method will allow us to reduce the time of production tests for the studied group of semiconductor devices intended for use in long-running equipment of critical purpose.