A human reliability model for numerically controlled machine centers.

摘要

Numerically controlled (NC) machining plays a vital role in today's automated industry. The control panel operation is the key job to conduct numerically controlled fabrication. To train a novice user, or re-train a current operator for an NC machine with a training simulator is a cost-effective strategy. The main purpose of this research is to develop a Human Reliability Model for NC control panel operation. In the field of reliability engineering, a great many of the studies have been conducted related to hardware and software reliability. Many databanks have been established related to electronic and mechanical components. As to human reliability, there is not so much empirical data except for some subjective data collected from experts. It is difficult to collect such data related to humans. An NC control panel system is a man-machine facility. Operation performance is not only affected by hardware and software in the system but also affected by the human software. For collection of errors of control panel operation, a web-based training simulator was modified from a previously used simulator by adding an operation data logger, and an automatic data recorder. The task, which was divided to five subtasks, of NC control panel operation is analyzed. Totally, 38 steps were involved in these five subtasks. An icon-based multimedia editor was used to develop the training simulator. It can be put on a web site that is located on the Internet or on the Intranet. A trainee could access the training simulator via the Internet or the Intranet. Data, operation error and performance, were automatically recorded. The error rate function was fitted to a power series function using the polynomial regression technique. The human reliability model for NC machining was developed based on reliability theory.;For application of the human reliability to training administration, a Java-based ARGEE system for reliability allocation was established. Using the system, individual reliability can be assigned to each subtask. This helps a training administrator to design courseware or training material for NC control panel operation. A feed-forward three-layered neural network was constructed for mapping target NC operation reliabilities and initial performances with training cycles, as well as breaks between two consecutive cycle and performances. Using the back-propagation algorithm, a Java-based system to train the weights of the neural network is explored. These network analysis techniques present training personnel in industry or educational institution with a tool for programming a training scheme and for evaluating training alternatives.