PreprintArticleVersion 1Preserved in Portico This version is not peer-reviewed
Preventing Forklift Front-End Failures: Predicting the Weight Center of Heavy Objects, Remaining Useful Life Prediction under Abnormal Conditions, and Failure Diagnosis based on Alarm Rules
Lee, J.-G.; Kim, Y.-S.; Lee, J.H. Preventing Forklift Front-End Failures: Predicting the Weight Centers of Heavy Objects, Remaining Useful Life Prediction under Abnormal Conditions, and Failure Diagnosis Based on Alarm Rules. Sensors2023, 23, 7706.
Lee, J.-G.; Kim, Y.-S.; Lee, J.H. Preventing Forklift Front-End Failures: Predicting the Weight Centers of Heavy Objects, Remaining Useful Life Prediction under Abnormal Conditions, and Failure Diagnosis Based on Alarm Rules. Sensors 2023, 23, 7706.
Lee, J.-G.; Kim, Y.-S.; Lee, J.H. Preventing Forklift Front-End Failures: Predicting the Weight Centers of Heavy Objects, Remaining Useful Life Prediction under Abnormal Conditions, and Failure Diagnosis Based on Alarm Rules. Sensors2023, 23, 7706.
Lee, J.-G.; Kim, Y.-S.; Lee, J.H. Preventing Forklift Front-End Failures: Predicting the Weight Centers of Heavy Objects, Remaining Useful Life Prediction under Abnormal Conditions, and Failure Diagnosis Based on Alarm Rules. Sensors 2023, 23, 7706.
Abstract
This study examined ways to prevent failures in the front-end of forklifts by addressing the center of gravity of heavy objects carried by forklifts, predicting the remaining useful lifetime (RUL), and the fault diagnosis based on alarm rules. In the research process, acceleration signals were acquired from the outer beam of the front-end of the forklift. A one-second window was applied to extract the time-domain statistical features, which were then set as variables. An exponentially weighted moving average was used to smooth the noise in the feature data set. The AWGN and LSTM autoencoders were used for data augmentation. Based on them, random forest and lightGBM models were used to develop classification models for the weight center of heavy objects carried by a forklift. In addition, contextual diagnosis performed by applying exponentially weighted moving averages to the classification probabilities of the machine learning models showed that random forest achieved an accuracy of 0.9563 and lightGBM achieved an accuracy of 0.9566. In addition, the acceleration data were collected through experiments to predict forklift failure and RUL because of the repeated forklift use when the center of heavy objects carried by the forklift was skewed to the right. The time-domain statistical features of the acceleration signals were extracted and set as variables by applying a 20-second window. Subsequently, logistic regression and random forest models were used to classify the failure stages of the forklifts. The f1-score (macro) obtained were 0.9790 and 0.9220 for logistic regression and random forest, respectively. In addition, the random forest probabilities for each stage were combined and averaged to generate a degradation curve and derive the failure threshold. The coefficient of the exponential function was calculated using the least squares method on the degradation curve, and the RUL prediction model was developed to predict the failure point. In addition, the SHAP algorithm was used to identify the significant features in classifying the stage. The alarm rule-based fault diagnosis was performed using the threshold of the normal stage distribution of the significant features.
Keywords
PHM; CBM; diagnosis; lightGBM; random forest; contextual diagnosis; RUL; forklift
Subject
Engineering, Mechanical Engineering
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
