PERANCANGAN SISTEM DETEKSI POSISI KERJA DENGAN RISIKO GANGGUAN OTOT RANGKA PADA PEKERJA KANTOR MENGGUNAKAN MACHINE LEARNING

• Main Author: Abdurrahim, Fauzan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/56964
• Institution: Institut Teknologi Bandung

Non-ergonomic work positions can affect worker performance and increase the risk of musculoskeletal disorders such as carpal tunnel syndrome, low back pain, and many more. An ergonomic workplace design alone will not suffice to establish an ergonomic work environment in the office. There must be supervision and monitoring of the most common work positions performed by employees, in this case the sitting position in front of the computer. Every day, office workers may spend 5-8 hours in front of a computer. The design of a work position detection system for office workers was carried out in this study. The cross-industry standard process for machine learning (CRISP-ML) was used to design the study approach. The process of developing a detecting system for office workers' work positions begins with data collecting from six participants consisting of three men and three women. Everyone was asked to work for two hours using a chair that is commonly used and a chair that is more ergonomic. A camera was used to record workers while they are at work. After that, 40 photos were taken randomly from each video and processed using OpenPose to get the coordinates of the joints and limbs of the workers. These coordinates was used to calculate the RULA and was used as predictor variables. The prediction model is created using two different algorithms, which are decision tree and random forest. The model is evaluated using three performance criteria, which are accuracy, precision, and recall. The best model is built with random forest algorithm. This model has a performance of 0.85 accuracy, 0.83 precision, and 0.76 recall. Following that, a flowchart was used to design the prototype design of the work position detecting system. The OpenPose program and prediction models generated using the random forest are integrated into the prototype design, allowing the assessment for office workers' work positions to be done in real-time.