A DCSLBP based intelligent machine malfunction detection model using sound signals for industrial automation systems

Abstract

Machine learning has an important role to create intelligent applications for Industry 4.0, and main objective of this paper is to contribute Industry 4.0 by using a sound signal dataset and a new feature engineering model to detect malfunctioning. This paper has evaluated a new sound classification model by using the comprehensive Malfunctioning Industrial Machine Investigation and Inspection (MIMII) dataset. The model has three stages: (i) feature generation with the newly developed Directed Center Symmetric Local Binary Pattern (DCSLBP) and Unbalanced Tree-Based Multilevel Discrete Wavelet Transform (UTMDWT), (ii) feature selection using the Chi-squared (Chi2) function, and (iii) classification with a Bagged Tree (BT) classifier. The MIMII dataset contains recordings of clean and noisy sounds at-6 dB and 6 dB from four types of machinery (fan, pump, slider, and valve), allowing for the creation of five specific test cases. Each case achieved classification accuracies >90%, demonstrating the effectiveness of the method. The results indicate that the proposed model, while achieving high accuracy in tests and providing a practical, less complex solution, is suitable for adaptations in real-world applications. Continuous improvements are essential due to the diverse industrial noise and operational conditions.