Low Back Exoskeletons in Industry 5.0: From Machines to Perceiving Co-Pilots—A State-of-the-Art Review

Sensors, Год журнала: 2025, Номер 25(7), С. 1958 - 1958

Опубликована: Март 21, 2025

This manuscript presents an updated review of back exoskeletons for occupational use, with a particular focus on sensor technology as key enabler intelligent and adaptive support. The study aims to identify barriers adoption explore design characteristics which align these systems the Industry 5.0 paradigm, where machines function collaborative co-pilots alongside humans. We propose structured pipeline analyze 32 across multiple dimensions, including design, actuation, control strategies, networks, intelligence. Additionally, we eight simulation environments support early stages exoskeleton development. Special emphasis is placed technology, highlighting its critical role in enhancing adaptability Our findings reveal that while 39.39% accommodate asymmetric activities, kinematic compatibility remains challenge. Furthermore, only 33.33% incorporated features, just one being capable adapting response based poor posture or real-time human-machine interaction feedback. limited integration advanced sensors decision-making capabilities constrains their potential dynamic Open questions remain high-level decision making, enhanced environmental awareness, development generalizable methods integrating data into strategies.

Язык: Английский

---

Machine Learning in Manufacturing Ergonomics: Recent Advances, Challenges, and Opportunities

IEEE Robotics and Automation Letters, Год журнала: 2021, Номер 6(3), С. 5745 - 5752

Опубликована: Июль 1, 2021

The rapid development of machine learning (ML) technology has introduced substantial impact on ergonomics research in manufacturing. Numerous studies and practices have been carried out to apply ML techniques address manufacturing issues, which brought extensive opportunities as well significant challenges. To incentivize future this area, letter reviews the recent advances applications ergonomics, discusses challenges from ML, systems perspectives.

Язык: Английский

---

Machine Learning-Based Fatigue Level Prediction for Exoskeleton-Assisted Trunk Flexion Tasks Using Wearable Sensors

Applied Sciences, Год журнала: 2024, Номер 14(11), С. 4563 - 4563

Опубликована: Май 26, 2024

Monitoring physical demands during task execution with exoskeletons can be instrumental in understanding their suitability for industrial tasks. This study aimed at developing a fatigue level prediction model Back-Support Industrial Exoskeletons (BSIEs) using wearable sensors. Fourteen participants performed set of intermittent trunk-flexion cycles consisting static, sustained, and dynamic activities, until they reached medium-high levels, while wearing BSIEs. Three classification algorithms, Support Vector Machine (SVM), Random Forest (RF), XGBoost (XGB), were implemented to predict perceived the back leg regions features from four wireless Electromyography (EMG) sensors integrated Inertial Measurement Units (IMUs). We examined best grouping sensor combinations by comparing performance. The findings showed performance binary 95% (2 EMG + IMU sensors) 82% (single sensor) accuracy, respectively. Tertiary required setups both measures perform 79% 67% efforts presented our article demonstrate feasibility an accessible detection system, which beneficial objective assessment, design selection, implementation BSIEs real-world scenarios.

Язык: Английский

---

A fatigue failure framework for the assessment of highly variable low back loading using inertial motion capture – a case study

Ergonomics, Год журнала: 2025, Номер unknown, С. 1 - 17

Опубликована: Фев. 16, 2025

Workers in manufacturing settings experience highly variable musculoskeletal loading, which current risk assessment methods often fail to fully capture. This study evaluated a Fatigue Failure-Based framework for estimating continuous lumbar loading from occupational loads. Worker movements and postures were recorded using Inertial Motion Capture technologies, L5/S1 joint history was estimated through inverse dynamics. Stress cycles analysed Rainflow analysis, adjusted with Goodman's method, summed Palmgren-Miner rule estimate cumulative damage. The tested live industrial eight automotive workers across 108 trials. Logistic regression models demonstrated significant correlations between damage estimates self-reported low-back pain (OR = 2.16, 95% CI: 1.30, 3.57). provides novel method analysing exposure ergonomics, offering starting point future research potential applications assessing low back injury risks similar settings.

Язык: Английский

---

Proof-of-concept system evaluation of Ergomechanic for non-invasive estimation of upper-body posture and body load exposure in the workplace

International Journal of Occupational Safety and Ergonomics, Год журнала: 2025, Номер unknown, С. 1 - 13

Опубликована: Фев. 24, 2025

Ergomechanic is a software-hardware system that uses cameras, computer vision and biomechanical modelling to calculate posture body load during physical activity in the workplace. This study evaluated its ability non-invasively automatically identify postures adopted by workers could lead injury long term, use these results suggest focuses for safety interventions. Five participants were recruited perform normal duties an area was view of four off-the-shelf security cameras. Participants randomly assigned work 60 h footage collected. metrics relating potential calculated each second footage. The extreme values metric used positions activities be hazardous term. Insights from process recommend changes design.

Язык: Английский

---

The perceptual and biomechanical effects of scaling back exosuit assistance to changing task demands

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 29, 2025

Back exoskeletons are gaining attention for preventing occupational back injuries, but they can disrupt movement, a burden that risks abandonment. Enhanced adaptability is proposed to mitigate burdens, perceptual benefits less known. This study investigates the and biomechanical impacts of SLACK suit (non-assistive) controller versus three controllers with varying adaptability: Weight-Direction-Angle adaptive (WDA-ADPT) scales assistance based on weight boxes using chest-mounted camera machine learning algorithm, movement direction, trunk flexion angle, standard Direction-Angle (DA-ADPT) Angle (A-ADPT) controllers. Fifteen participants performed variable (2, 8, 14 kg) box-transfer task. WDA-ADPT achieved highest score (88%) across survey categories reduced peak extensor (BE) muscle amplitudes by 10.1%. DA-ADPT had slightly lower (76%) BE reduction (8.5%). A-ADPT induced hip restriction, which could explain lowest (55%) despite providing largest reductions in activity (17.3%). Reduced scores DA were explained too much or little actual task demands. These findings underscore scaling demands improves perception device's suitability.

Язык: Английский

---

Wearable sensors for classification of load-handling tasks with machine learning algorithms in occupational safety and health: a systematic literature review

Ergonomics, Год журнала: 2025, Номер unknown, С. 1 - 21

Опубликована: Апрель 17, 2025

Ergonomic assessments are critical to preventing work-related musculoskeletal disorders. The integration of machine learning with wearable sensor technology offers new approaches risk assessment by capturing external forces and non-ergonomic working conditions. We conducted a systematic literature search, reviewing 851 studies from PubMed, Web Science Embase, included 15 in our analysis. This review summarises critically discusses these studies, which focus on posture classification, activity duration weight estimation load handling tasks. Although the results promising, current research covers only few aspects, limited emphasis measurement forces. Furthermore, many faced fundamental issues such as small sample sizes access data algorithms. Future advances this area could greatly benefit sharing datasets algorithms, thereby increasing comparability robustness findings.

Язык: Английский

---

Real-Time Ground Reaction Force and Knee Extension Moment Estimation During Drop Landings Via Modular LSTM Modeling and Wearable IMUs

IEEE Journal of Biomedical and Health Informatics, Год журнала: 2023, Номер 27(7), С. 3222 - 3233

Опубликована: Апрель 27, 2023

This work investigates real-time estimation of vertical ground reaction force (vGRF) and external knee extension moment (KEM) during single- double-leg drop landings via wearable inertial measurement units (IMUs) machine learning. A real-time, modular LSTM model with four sub-deep neural networks was developed to estimate vGRF KEM. Sixteen subjects wore eight IMUs on the chest, waist, right left thighs, shanks, feet performed landing trials. Ground embedded plates an optical motion capture system were used for training evaluation. During single-leg landings, accuracy KEM R2 = 0.88 ± 0.12 0.84 0.14, respectively, 0.85 0.11 0.12, respectively. The best estimations optimal unit number (130) require placed selected locations landings. a leg only needs five leg's shank, thigh, foot. proposed LSTM-based optimally-configurable can accurately in relatively low computational cost tasks. investigation could potentially enable in-field, non-contact anterior cruciate ligament injury risk screening intervention programs.

Язык: Английский

---

Machine learning approach to determine the decision rules in ergonomic assessment of working posture in sewing machine operators

Journal of Safety Research, Год журнала: 2023, Номер 87, С. 15 - 26

Опубликована: Авг. 22, 2023

Язык: Английский

---

Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles

Sensors, Год журнала: 2024, Номер 24(16), С. 5318 - 5318

Опубликована: Авг. 16, 2024

The three Ground Reaction Force (GRF) components can be estimated using pressure insole sensors. In this paper, we compare the accuracy of estimating GRF for both feet six methods: Deep Learning (DL) methods (Artificial Neural Network, Long Short-Term Memory, and Convolutional Network) Supervised Machine (SML) (Least Squares, Support Vector Regression, Random Forest (RF)). Data were collected from nine subjects across activities: normal slow walking, static with without carrying a load, two Manual Material Handling activities. This study has main contributions: first, estimation (Fx, Fy, Fz) during activities, which have never been studied; second, comparison component between each activity. RF provided most accurate situations, mean RMSE values RMSE_Fx = 1.65 N, RMSE_Fy 1.35 RMSE_Fz 7.97 N absolute measured by force plate (reference) 14.10 3.83 397.45 N. our study, found that RF, an SML method, surpassed experimented DL methods.

Язык: Английский

---