Copyright © All rights reserved micansinfotech.com
Industrial safety has been deeply improved in the past years, thanks to increasingly sophisticated technologies. Neverthe- less,2.3millionpeopleyearlydieworldwideduetooccupationalill- nessesandaccidentsatwork.Humanfactorscanbeprofitablyused for safety improvement because of their influence on the workers’ behavior. This paper presents an integrated optimization system to help companies assign each task to the most suitable worker, minimizing cost, while maximizing expertise and safety. The sys- tem is made of three modules. A neural module computes each worker’s caution for every task on the basis of some human factors and the worker’s behavior. To solve the multiobjective job assign- ment problem, an evolutionary module approximates the Pareto front through the nondominated sorting genetic algorithm II. Pareto-optimal solutions then form the alternatives of a multi- criteria decision-making problem, and the best is selected by a decision module jointly based on the analytic hierarchy process and the technique for order of preference by similarity to ideal solution. Validation was carried out involving two footwear com- panies, where personnel was recruited and reassigned to tasks, respectively. Comparing the worker-task assignment proposed by the system to the one suggested/used by the management, note- worthy low-cost improvement in safety is shown in both scenarios, with low or no decrease in expertise. The proposed system can, thus, contribute to get safer workplaces where risks are less likely and/or less harmful.
The neural module determines each worker’s risk perception and caution levels with respect to the risks of every single task, taking as inputs the worker’s values for the human factors and his/her behavior while executing the task. The worker’s behav- ior is represented by means of one or more preventive actions that the worker performs or would perform while executing the task. Each worker’s level of safety in performing every task is determined by aggregating his/her risk perception and caution levels toward that task.
randomly generating a set ofcandidate solutions to the problem, forming the initial population. Each individual’s goodnessismeasuredbyusingafitnessfunction:e.g.,foramini- mizationproblem,thelowerthefitness,thebettertheindividual. Individuals with good fitness are more likely to be selected for reproduction, which takes place by means of crossover and/or mutation operators. A new population (offspring) is created by replacing (part of) the individuals of the current population with the newly generated ones. The process iterates until a terminat- ingconditionismet.
Semisupervised learning within a stage-based learning scheme was used. A supervised learning stage first trains the MLPs separately. Then, each MLP’s performance is improved, thanks to what the other learned previously. To this aim, un- supervised data are used to generate, through each MLP, the desired outputs from the other. This is possible because the two MLPs, which receive as inputs two different representations of the same person, should produce coherent outputs. The so- generated training sets are used to refine the training of the two MLPsstartingfromthevaluesassumedbytheneuralparameters at the end of the previous stage.
The procedure is repeated until chromo- somes are over. At iteration t, an offspring population Q t of n individuals is generated by selecting from current population P t and, then, performing crossover and mutation. A new popula- tion R t of 2n chromosomes is generated merging P t and Q t . Chromosomes in R t are assigned to their ranks; hence, R t is partitioned into fronts. For each front, the density of individuals in each individual’s neighborhood is estimated with the crowd- ing distance, i.e., the sum of the distances from an individual to the closest one, along each objective. code.
A genetic algorithm (GA) is a heuristic optimization method based on biological evolution [22]. GAs efficiently deal with complex single-objective op- timization or MOO problems. Solutions, i.e., individuals, are encoded by using bit strings,integer- or real-valued vectors, etc. Eachindividualistypicallymadeofonechromosomecomposed of genes.y8
Humanfactorscanbeprofitablyused for safety improvement because of their influence on the workers’ behavior. This paper presents an integrated optimization system to help companies assign each task to the most suitable worker, minimizing cost, while maximizing expertise and safety. The sys- tem is made of three modules. A neural module computes each worker’s caution for every task on the basis of some human factors and the worker’s behavior
Risk programs include risk awareness training, which is generally iterated to stress the concepts so that learning is reinforced, with the aim of an ever-increasing injury reduction. This is quite expensive for companies. The training outcome can also be enhanced by us- ing training methods specifically tailored to every single worker [14], exploiting his/her sensitivity to risk
To solve the multiobjective job assign- ment problem, an evolutionary module approximates the Pareto front through the nondominated sorting genetic algorithm II. Pareto-optimal solutions then form the alternatives of a multi- criteria decision-making problem, and the best is selected by a decision module jointly based on the analytic hierarchy process and the technique for order of preference by similarity to ideal solution. Validation was carried out involving two footwear com- panies, where personnel was recruited and reassigned to tasks, respectively.
pareto-optimal solutions then form the alternatives of a multi- criteria decision-making problem, and the best is selected by a decision module jointly based on the analytic hierarchy process and the technique for order of preference by similarity to ideal solution. Validation was carried out involving two footwear com- panies, where personnel was recruited and reassigned to tasks, respectively..
This paper presented an integrated optimization system, which helps the management of a company assign workers to tasks minimizing cost and maximizing expertise and safety. The system is made of three modules. A neural module de- termines how safely a worker can be assigned to a task by using some worker’s human factors and the accuracy of the preventive actions performed during the task execution.
The multiobjective job assignment problem is solved by a hybrid evolutionary/MCDM resolution methodology. The evolution- ary module generates an accurate approximation of the Pareto front by means of the NSGA-II. An MCDM technique based on AHP and TOPSIS is used by the decision module to select the near Pareto-optimal solution representing the best compromise of the preferences expressed by the management.