APPLICATION OF RIDGE REGRESSION FOR DIESEL ENGINE ABNORMALITY DETECTION

• Main Author: Khairul Arifin, Fariz
• Format: Theses
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/52256
• Institution: Institut Teknologi Bandung

Every diesel engine manufacturer usually has set the limits for the value of the operation variables where the engine can be considered as running normally. However, the abnormal symptoms can be detected before the failure occurs, even though the value of the operating variable does not exceed its limit. To prevent further damage, action can be taken as soon as possible after the abnormality symptoms detected. In this study, a method for detecting abnormal symptoms was developed based on ridge regression (RR) with Hoerl and Kennard (HK) method for bias determination combined with Q statistic and variable reconstruction. Ridge regression is used to predict the values of operating variables, while Q statistic and variable reconstruction are used to detec abnormal symptoms based on the prediction results. In this study, the regression model vas varied into three, namely conventional (HK), recursive (HKR), and moving window (HKMW). The Q statistic detection method is varied into two, namely per observation Q statistic (method A) and per variable Q statistic (method B). in addition, this study also discussed the effect of the variable reconstruction (R) on the abnormalities detection results. The combination of the above variations produces 12 methods: HK_A, HKR_A, HKMW_A, HK_AR, HKR_AR, HKMW_AR, HK_B, HKR_B, HKMW_B, HK_BR, HKR_BR, and HKMW_BR. The best combination is determined by comparing the detection results obtained from each combination with the maintenance history data based on the similarity level to the maintenance history and the pattern of the number of detections. The results obtained indicate that HKMW_BR method has the highest similarity level compared to other methods that is 27 out of 36 maintenance activities detected based on their abnormal symptoms. In addition, the HKMW_BR method also gives the best detection results based on the pattern of the number of abnormal variables detected: 69 out of 76 maintenance activities detected.