Techniques based on adaptive neuro-fuzzy inference systems (ANFIS) for estimating and evaluating physical demands at work using heart rate.

摘要

Despite the rapid evolution of mechanization in heavy industries, physically demanding jobs that require excessive human effort still represent a significant part of many industries (e.g., forestry, construction, mining etc.). Studies have shown that excessive workloads placed on workers are the main cause of physical fatigue, which has negative effects on the workers, their performance and quality of work. Therefore, researchers have emphasized on the importance of the optimal job design (within workers' capacity) in order to maintain workers' safety, health and productivity. However, this cannot be achieved without understanding (i.e., measuring and evaluating) the physiological demands of work. In this respect, the three studies comprising this dissertation present practical approaches for estimating and evaluating energy expenditure (EE), expressed in terms of oxygen consumption (VO2), during actual work.;The first study presents new approaches based on adaptive neuro-fuzzy inference system (ANFIS) for the estimation of VO2 from heart rate (HR) measurements. This study comprises two stages in which 35 healthy individuals participated. In the first stage, two novel individual models were developed based on the ANFIS and the analytical methods. These models tackle the problem of uncertainty and nonlinearity between HR and VO2. In the second stage, a General ANFIS model was developed which does not require individual calibration. The three models were tested under laboratory and field conditions. Performance of each model was evaluated and compared to the measured VO 2 and two traditional individual VO2 estimation methods (linear calibration and Flex-HR). Results indicated the superior precision achieved with individualized ANFIS modeling (RMSE= 1.0 and 2.8 ml/kg.min in laboratory and field, respectively). The analytical model outperformed the traditional linear calibration and Flex-HR methods with field data. The General ANFIS model's estimates of VO2 were not significantly different from actual field VO2 measurements (RMSE= 3.5 ml/kg.min). With its ease of use and low implementation cost, the General ANFIS model shows potential to replace any of the traditional individualized methods for VO2 estimation from HR data collected in the field.;The second study presents an ANFIS-based VO2 prediction model that is inspired by the Flex-HR method. Studies have shown that the Flex-HR method is one of the most accurate methods for VO2 estimation. However, this method is based on four parameters that are determined individually and therefore it is considered costly, time consuming and often impractical, especially when the number of workers increases. The proposed prediction model consists of three ANFIS modules for estimating the Flex-HR parameters. For each ANFIS module, input variables selection and model assessment were simultaneously performed using the combination of three-way data split and cross-validation techniques. The performance of each ANFIS module was tested and compared with the observed parameters as well as with Rennie et al.'s (2001) models using independent test data. In addition, the performance of the overall ANFIS prediction model in estimating VO2 was tested and compared with the measured VO2 values, the standard Flex-HR method as well as with other general models (i.e., Rennie et al.'s (2001) and Keytel et al.'s (2005) models). Results indicated no significant difference between observed and estimated Flex-HR parameters and between measured and estimated VO2 in the overall HR range, and separately in different HR ranges. The ANFIS prediction model (MAE = 3 ml/kg.min) demonstrated better performance than Rennie et al.'s (MAE = 7 ml/kg.min) and Keytel et al.'s (MAE = 6 ml/kg.min) models, and comparable performance with the standard Flex-HR method (MAE = 2.3 ml/kg.min) throughout the HR range. The ANFIS model thus provides practitioners with a practical, cost- and time-efficient method for VO2 estimation without the need for individual calibration.;The third study presents a new approach based ANFIS for classifying work intensity into four classes (i.e., very light, light, moderate and heavy) by using heart rate monitoring. Intersubject variability (physiological and physical differences) was considered. Twenty-eight participants performed Meyer and Flenghi (1995) step-test and a maximal treadmill test, during which heart rate and oxygen consumption were measured. Results indicated that heart rate monitoring (HR, HRmax, and HRrest) and body weight are significant variables for classifying work rate. The ANFIS classifier showed superior sensitivity, specificity, and accuracy compared to current practice using established work rate categories based on percent heart rate reserve (%HRR), with an overall 29.6% difference in classification accuracy between the two methods, and good balance between sensitivity (90.7%, on average) and specificity (95.2%, on average). With its ease of implementation and variable measurement, the ANFIS classifier shows potential for widespread use by practitioners for work rate assessment.