Process control teaks typically require the operator to diagnose and rectify failures on tho basis of parameters measured at different points in the plant and centrally displayed on a control panel. An i nvoat.igat.ion was carried out to discover what form of technical introduction would be most successful in training an operator accurately to recognize faults practised during training, and in addition provide a strategy enabling tho trainee to Book out relevant information in order to diagnose unfamiliar faults.Throe groups of subjects were trained using an adaptive cumulative-part technique to recognize eight fuulte displayed on a simulated control panel. Each group received a different introduction to the plant: (i) the t Theory' group were given it conventional description of the t Plant' and its functions; (ii} tho' Rules I group in addition were taught a sot of rules that would assist them in inferring failures from the panel array; (iii) the' No Story' group underwent training without any prior theoretical i.ntroduction.After cumulative-part training ell three groups performed equally well on the eight faults. However, the' Rules ' group were consistently the most proficient at diagnosing' unfamiliar' faults. Although tile' Theory' group initially were more able at recognizing' unfamiliar' faults, after intensive training their performance bud deteriorated to a level not significantly different from the I No Story' group.It is argued, therefore, that in an effective traini.ng regime, tho teaching of conventional 'Theory' alone is of limited value, and for versatile diagnostic performance, training should include generalizable rules of the kind described in this paper.

The development of a training course on fault finding in a crude distillation unit is described and its contents illustrated. The criterion problem for this and similar industrial problem‐solving tasks is discussed. ‘Operational’, or product measures, namely, first shot successes (FSS), wrong diagnoses (WD) and diagnosis time (T), are reported and compared with five measures of the problem‐solving process. Training substantially increased FSS, substantially reduced T and reduced WD by an order of magnitude. Post‐training scores were superior to those of men with refinery operating experience, although the difference in T was not statistically significant. Large and significant improvements were also found with the majority of the process measures. The training course only required five sessions of 2½ hours including pre‐ and post‐tests.

The effects were studied of two types of training of fault-location performance in networks of interconnected logic units. The types of training were (i) computer aiding which provided information about the efficiency of tests and diagnoses not warranted by available indications, and (ii) pre-training in strategies appropriate to different network configurations. Efficiency of testing was more easily improved than errors of diamosis, especially by a combination of both pre-training and computer aiding. Improvements in diagnostic quality can apparently be achieved, although the effects of pre-training atone may be short lived. It seems that computer aiding is most effective when the subject has first obtained some understanding of the problem, and is thus in a better position to make use of the information about the quality of his problem solving.

It has been suggested that complexity in fault diagnosis tasks is largely a function of problem complexity and little to do with the perceptual complexity of the format in which the system representation is displayed. An experimental study is reported which shows that a left-to-right diagonal display format improves the speed and diagnostic efficiency with which faults are located. The differences in display format which appear to be most beneficial do not affect problem-solving complexity in any obvious way. Differences in display format which could affect problem-solving complexity, such as constraints on direction of signal flow, do not apparently affect fault diagnosis performance. The improvements in performance apparently stem from an increased ability to perceive components of the system relevant to the observed symptoms.

No abstract