Publicação
Work-related Upper Body Postures Classification and Segmentation
| Resumo: | The European Union faces a significant challenge, with three out of five industrial workers reporting musculoskeletal injuries linked to work, primarily affecting the upper limbs due to extreme postures, repetitive movements, and heavy object handling. Industry 4.0 and 5.0 have prompted a digital transformation in the industrial sector, emphasizing sustainable productivity and worker’s well-being. In response, the OPERATOR project aims to enhance work ergonomics evaluation, fostering awareness of worker-workplace interaction to mitigate potential health issues. This project involves developing an automatic ergonomic assessment for Autoeuropa’s automotive assembly line, using inertial measurement units (IMUs). This automation extends to the European Assembly Worksheet (EAWS) and integrates International Organisation for Standardization (ISO) norm 11226 standards to support ergonomists decision-making. The present dissertation addresses two key challenges: classifying complex postures and developing a dashboard for ergonomic practice support. A deep learning model was developed to perform posture classification focusing on complex working postures, such as arms at shoulder level or overhead work. Using the AnDy dataset, which is composed of automotive industry-like tasks collected by 17 IMUs and annotated according to the EAWS, the model achieved recall values of 0.95 and 0.97 for shoulder-level and overhead working postures, respectively. The two mentioned postures are fed into ergonomic exposure and risk determination algorithms, utilizing ISO 11226 and EAWS. Additionally, a dashboard was co-developed with two Autoeuropa ergonomists aiding with the laborious and time-consuming risk assessment currently performed. The tool integrates data from the ergonomic exposure and risk algorithms, presenting graphs and metrics. It calculates risk exposure percentages and number of occurrences, as well as provides risk scores for five work postures according to the EAWS, ultimately enabling ergonomists to perform a more detailed, simple and personalised ergonomic risk assessment for each worker. |
|---|---|
| Autores principais: | Sózinho, Diogo Canadas |
| Assunto: | Lesões musculoesqueléticas relacionadas com o trabalho Reconhecimento de posturas Análise de séries temporais Ergonomia ocupacional Teses de mestrado - 2024 |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | The European Union faces a significant challenge, with three out of five industrial workers reporting musculoskeletal injuries linked to work, primarily affecting the upper limbs due to extreme postures, repetitive movements, and heavy object handling. Industry 4.0 and 5.0 have prompted a digital transformation in the industrial sector, emphasizing sustainable productivity and worker’s well-being. In response, the OPERATOR project aims to enhance work ergonomics evaluation, fostering awareness of worker-workplace interaction to mitigate potential health issues. This project involves developing an automatic ergonomic assessment for Autoeuropa’s automotive assembly line, using inertial measurement units (IMUs). This automation extends to the European Assembly Worksheet (EAWS) and integrates International Organisation for Standardization (ISO) norm 11226 standards to support ergonomists decision-making. The present dissertation addresses two key challenges: classifying complex postures and developing a dashboard for ergonomic practice support. A deep learning model was developed to perform posture classification focusing on complex working postures, such as arms at shoulder level or overhead work. Using the AnDy dataset, which is composed of automotive industry-like tasks collected by 17 IMUs and annotated according to the EAWS, the model achieved recall values of 0.95 and 0.97 for shoulder-level and overhead working postures, respectively. The two mentioned postures are fed into ergonomic exposure and risk determination algorithms, utilizing ISO 11226 and EAWS. Additionally, a dashboard was co-developed with two Autoeuropa ergonomists aiding with the laborious and time-consuming risk assessment currently performed. The tool integrates data from the ergonomic exposure and risk algorithms, presenting graphs and metrics. It calculates risk exposure percentages and number of occurrences, as well as provides risk scores for five work postures according to the EAWS, ultimately enabling ergonomists to perform a more detailed, simple and personalised ergonomic risk assessment for each worker. |
|---|