휴먼에러를 고려한 크레인 작업자의 신뢰성 향상에 관한 연구

Abstract

The accidents such as at Chernobyl and Bhopal have demonstrated unequivocally the importance of considering human error in higher risk systems. For any existing plant, or new one being designed, it is importance to try to assess the likelihood of such accidents and prevent them from occurring. This requires the assessment of the impact of human errors on system safety and, if warranted, the specification of ways to reduce human error impact and/or frequency. For generic approach for assessment of human error, there are three goals, namely: the human error identification is first step to reduce human error in higher risk systems. The second step can provide quantification of error which might be needed to construct a safety case. The final step will develop error data-bases to reduce human error.

Interest in modeling the behavior of a human as an active feedback control device began during Word War II, when engineers and psychologists attempted to improve the performance of pilots, gunners, and bombardiers. To design satisfactory manually controlled systems many researchers began analyzing the neuro-muscular characteristics of the human operator. Their approach was to consider the human as an inanimate servomechanism with a well-defined input and output.

Over the years, the evolution of the control-theory paradigm for the human controller or operator paralleled the development of new synthesis techniques in feedback control. Thus, "optimal control models"(OCMs) of the human operator appeared as linear quadratic Gaussian (LQG) control system design techniques were being developed. "Fuzzy controller" models and "H infinity" models of the human operator closely followed the appearance of these design techniques.

Recently, a human operator (HO) dynamics have been proposed to describe the human operation model in manual tracking tasks. The model has been derived through the application of classical and modern control theory or time-series analysis. Structural isomorphic models were the result of applying classical control theory. These models seek to account for many of subsystem characteristics of the human operator by assigning transfer functions to the different subsystems involved. These subsystems and their interconnections are postulated on the basis of physiologically isomorphic considerations.

Especially, the application of the time-series analysis to this problem was first introduced by Shinners who developed autoregressive moving-average (ARMA) models of data collected from human operators involved in compensatory tracking experiments using band limited white noise inputs (bandwidth=1.5Hz). In this result, all operators exhibited a time delay 0.2s and the discrete transfer functions that represent their dynamics had one zero and two poles. Based on the analysis of model residuals, he concluded that the human operator is a generator of seasonal (rhythmic) characteristics during tracking of random inputs.

On the other hand, Charles 1980 and Malek 1988 developed a transfer function model from input-output data collected from a HO during both compensatory and pursuit manual tracking experiments. In their experiment, two unpredictable inputs, formed by the addition of five sinusoids, were used. The first had a low-frequency range (0.04-0.8 Hz) and the second had a high-frequency range (0.08-1.48 Hz). For the low frequency range, they were able to fit a transfer function with two poles and no zeros, while a transfer function with two poles and one zero adequately fitted the high-frequency range input-output data. Analysis of model residuals showed no sign of rhythmic characteristics that were observed earlier by Shinners.

In this thesis, we will apply to input-output data of human operator involved in transport of container by using the time-series analysis, specifically ARX modeling. Our aim is to provide a simple model which use to quantify variations in the HO dynamical behavior in transport of container following given paths, this is a kind of manual tracking tasks. The input data of human operator are reference path, and the state of gantry crane (the position of trolley, the sway angle, and the length of cable

and also the first order and the second order of differentiating of these states). The output data of human operator are two angles in horizontal and vertical handles on the joystick which proportionate with the forces on trolley and cable, respectively.

First, we propose a human model for analysis for human work pattern or human fault, where a gantry crane simulator is used to survey the property of human operation. From the input and output of gantry crane response, we make a human operation model by using conventional ARX identification method. For identify the human model, we assume the six inputs and two outputs. By using the input/output data, we estimate the parameters of ARX of the human system model. For verify the proposed method, we compared the real data with the modeled data, where three kinds of work trajectory path are used.

Second, we deal with an observer design for detecting the human faults in container crane operation, where an observer for detecting the human faults was proposed and the existing condition for the observer was shown. In this case, we assume that the human faults can be considered as a careless mistake during the crane operation. In simulation, we used the previous results for human work model and design the observer for the human work model. As a simulation results with human faults, the proposed observer can detect the human faults perfectly, thus the efficiency of proposed observer is shown.

Lastly, we proposed an reliable improvement procedure for control the gantry crane with human operation. In here, the control input for gantry is compensated by human errors, where the human errors are detected and isolated by previous observer design results. In simulation, we verify that the constructed reliable control procedure for gantry crane system can improve the total human-machine operation system.