Risk Assessment And Prediction Of Unsafe Behavior In Machinery Manufacturing Industry

As a traditional manufacturing industry,mechanical processing and manufacturing industry has received less attention in safety studies.However,some studies have shown that assembly line workers are more prone to unsafe behaviors due to their working nature.In addition,according to the accident causation theory,both the unsafe behavior of people and the unsafe state of things can lead to the occurrence of accidents,but a large number of scholars’ studies show that at least 80%of the direct causes of accidents are attributed to the unsafe behavior of people.Therefore,it is necessary to evaluate and predict the risks of unsafe behaviors of employees in the mechanical processing and manufacturing industry,and put forward reasonable measures and control suggestions,so as to reduce the probability of unsafe behaviors and help enterprises achieve good safety management results.First,the classification criteria,influencing factors,accident risk assessment and prediction methods of unsafe behaviors were studied by combing relevant domestic and foreign literatures.The contents and methods of the research are defined,that is,the unsafe behavior is classified through the Classification Standard of Injuries and Casualties of Enterprise Employees(GB6441-86),the index system of the influencing factors of the unsafe behavior is roughly determined,and the risk of the unsafe behavior is evaluated and predicted through weighted fuzzy logic,Monte Carlo method and neural network.Second,the index system of the influencing factors of unsafe behaviors in the HSE Guide is selected based on the actual situation of A Company and the literature review.Secondly,the data of accidents directly caused by unsafe behaviors in the last 5 years are classified according to the time series,and the weights of the secondary and tertiary indexes in the index system are calculated by the grey correlation method.Next,according to the literature integration,the language variables,domain,fuzzy set,membership function,fuzzy rules and the evaluation questionnaire of the influencing factors of unsafe behavior in the weighted fuzzy theory are determined,and the adjustment is made by referring to the expert opinions.Finally,the senior management of the enterprise is investigated through the questionnaire,and the score of the questionnaire is input into the weighted fuzzy logic,and the probability correction coefficient of the unsafe behavior within the enterprise is finally obtained.Third,the risk assessment function of unsafe behavior is constructed firstly.Secondly,the unsafe behavior of accident data is classified according to the Classification Standard of Enterprise Employee Casualty Accidents,and the probability distribution of the severity of each accident including the attempted event is calculated.Then Monte Carlo method was used to simulate the probability distribution of the risk value of accidents caused by unsafe behaviors in each working position.Combined with the risk matrix,the risk level was obtained and countermeasures and suggestions were put forward.Fourth,in order to reduce the complexity of weighted fuzzy logic in the actual operation,neural network is used to predict the unsafe probability correction coefficient.Firstly,fuzzy logic was used to calculate the probability modifications of 77 accidents,which were divided into 58 training samples and 19 test samples.Secondly,the samples were fed into the BP neural network according to the third-level index as the input layer and the second-level index as the input layer,and the comparison results showed that the second-level index as the input layer had better effect.Finally,the T-S adaptive fuzzy neural network is fed with the second-level indexes as the input layer,and the fitting results are compared with the BP neural network using the second-level indexes as the input layer.It is concluded that the T-S adaptive fuzzy neural network shows better effect than the BP neural network.The second-order index values measured in Chapter 3 are input to T-S adaptive fuzzy neural network for verification,and the predicted results are almost identical with the actual values calculated in Chapter 3.Therefore,the neural network model can be considered reasonable and effective.