A Probabilistic Method for Constructing an Empirical Discrimination Model for Hammering Inspection of Cast-Iron Parts

Abstract

A probabilistic method for constructing an empirical discrimination model to inspect defective cast-iron parts (such as graphite-spheroidized defective parts) by the hammering test is proposed. The hammering-sound frequency spectrum includes multiple resonance lines whose frequencies vary according to the degree of defect. To construct the model, only non-defective hammering-sound data that can be collected from the production line are input, and a distribution function that fits the frequency distribution of each resonance line is estimated. Since the frequency distribution shows multimodality and asymmetry, the function is estimated by using automatic differentiation variational inference with a mixed-normal distribution function. The confidence interval of the obtained distribution function is then regarded as a section with no defective parts, and the discrimination model is automatically constructed by connecting the sections of all resonance lines in the audible range. Then, parts outside the sections are discriminated as defective. Experimentally determined accuracy confirmed that it is possible to achieve hammer-test inspection with the detection rate of 100% and prevent overlooking of defective parts.

Keywords

• hammer test
• cast iron
• statistical model
• automatic differentiation variational inference
• markov chain monte carlo methods