This paper provides a novel approach for condition-based monitoring based on lubricant analysis and machine-learning algorithm. We collected used lubricant oil samples from marine diesel engines. Samples were analyzed by using X-ray Fluorescence analysis. Wear elements such as iron were detected quantitatively. The relationship between elements were visualized by Gaussian Graphical Model(GGM), and it was suggested that iron particles mainly come from wear modes such as the contact between cylinder and piston ring. Kullback-Leibler divergence was used to quantify the change in the GGM. Anomaly scores were calculated for samples and the scores depends on ship routes. Sensitivity analysis revealed that sulfur and calcium from fuel affect the increase in the anomaly score. This approach may provide the quantitative novel information to condition based maintenance.
The industrial accidents concerning gas cutting and welding had occurred frequently. The research institute (JNIOSH) collected the accident information and conducted a field survey. Gas cutting apparatus were collected from many factories and measured their performance. The collection work was done with the cooperation of the Japan Welding Engineering Society (JWES). In this paper, only the results of the dry flashback arrestor are described. 79 dry flashback arrestors for fuel and 31 for oxygen were collected from 50 factories. The collected arrestors had been used for 3 to 21 years. We analyzed the collection questionnaire and measured the performance of those arrestors. In the performance measurement, the visual inspection, the gas-leak test, the reverse-flow test, and the cut-off test were performed. Furthermore, deteriorated, or defective arrestors were disassembled, and the inside was examined. Failures are seen in 3 to 4 years. It is found that the largest number of failures is revers-flow for both fuel and oxygen.