Sound-Based Fault Detection For Textile Machinery.

The textile sector is one of the vital driving forces in the economy of south Asian countries like Bangladesh, India, Pakistan, etc. However, most of the textile industries suffer from frequent machinery faults everyday which reduces their productivity, which in terms reduces their profit. Existing systems for detecting the faults in textile machinery fails to find a remedy to this problem due to several limitations. Among them, sound based and vibration based fault detection systems are based on prototype machinery and has smaller data set to detect machinery fault properly. The fabric defect based, machine learning based approaches only detect machinery fault after fabric has become already defected. To remedy these limitations, in this paper, we propose a sound based fault detection system consisting of trained machine learning model from large data set that can detect machinery fault in textile industry. We use a sound sensor to measure the sound signal of the machine. We artificially create three real faults in the experimented machine and measure the sound signal during the faults. Next, we conduct Fast Fourier Analysis derive sound frequency and statistical analysis to derive different statistical features from the prepared data set. From these two analysis, we determine if the sound frequency and amplitude changes during the fault. After that, we feed the data set to ten machine learning algorithms. Finally, we evaluate our trained machine leaning models through ten fold cross validation to determine the precision, recall, and F1 score. We find the highest F1 of 57.7% in Nearest Centroid Algorithm.