This paper presents the development of a procedure for the objective, repeatable, and automated final inspection of glow plugs with a built-in pressure sensor. The process is based on an analysis of the response of a piezoelectric sensor element during calibration, during which vibrations from the calibration device cause forced oscillations in the sensor assembly. The amplitude of this oscillation depends on the mechanical properties and quality of the glow plug’s construction, thus providing a useful basis for product classification. To gain insight into the frequency-specific characteristics of the measured signals, a Fourier transform was performed. For effective machine learning, the datasets were subsequently further processed using various feature extraction approaches. Various machine learning algorithms were trained on the resulting dataset to classify the glow plugs into quality classes. Cross-validation was used to test various combinations of feature extraction approaches and machine learning algorithms. The resulting evaluation metric values served as the basis for identifying the most successful combination. The developed procedure thus provides a foundation for improving the reliability and reproducibility of end-of-line control.
|
