DEVELOPMENT OF FATIGUE PREDICTION MODELS IN MANUAL MATERIAL HANDLING WORK USING MACHINE LEARNING

• Main Author: Abdurrahim, Fauzan
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/69350
• Institution: Institut Teknologi Bandung

Physical fatigue is a very common phenomenon in industry. Manual material handling (MMH) activities, which include pushing, pulling, and carrying materials, are one of the work activities associated with physical fatigue. Physical fatigue can be measured physiologically and biomechanically. Biomechanical fatigue can be measured by computing biomechanical parameters such as kinetic and kinematic parameters. Physical fatigue measurement using a biomechanical approach, which is currently being pursued, necessitates a lengthy processing time, beginning with data collection and ending with measurement results. Machine learning can be used to detect physical fatigue quickly and in real-time, which could be an alternative method of measuring physical fatigue in the workplace. In this study, a machine learning model was developed to detect physical fatigue using various biomechanical parameters. The prediction model design process begins with data collection from 12 male participants ranging in age from 20 to 30 years and body mass index (BMI) from 18 to 30kg/m2. Everyone will be required to complete a series of manual material handling (MMH) activities. The fatigue protocol is to run for 8 minutes at a speed of 8-12km/h. The fatigue was then assessed using Borg's rating of perceived exertion scale (RPE). If the Borg’s RPE value is greater than 17, the participant is considered tired. Participants will be asked to run again if they do not comply. Vicon Nexus motion capture system ® and AMTI force platform systems ® are used in the data retrieval process. In the prediction model, three kinetic parameters and sixteen kinematic parameters were used as predictors. The machine learning model development process is then carried out. The kinetic parameters analysis showed that the value of the ground reaction force (GRF) in the fatigue condition differed significantly from the non-fatigue condition in several phases. The GRF value is higher in the fatigue condition than in the non-fatigue condition. The parameters of the right and left knee joint velocity, right and left knee joint acceleration, and right ankle joint velocity on the y-axis are significantly different when the average value of fatigue and non-fatigue conditions are compared. Although there were no significant differences in the other parameters, there was a trend in which the mean parameter was higher in the fatigued condition than in the non-fatigued condition. The random forest algorithm was used to create three predicted models. The prediction model chosen is one with kinematics parameter predictors. The chosen model has 98% accuracy, 98% precision, and 98% sensitivity. In each prediction model, the importance value of parameter is balanced. As a result, all of the biomechanical parameters used in this study can be used as fatigue indicators.