group. In the phase of fault location, a solution giving as faulty a component of the testable group uniquely locates the fault, while a solution giving as faulty one of the components of the two global ambiguity groups indicates only that there is a fault in that group, without allowing the determination of the effectively faulty component. It is worth noting that these results are theoretical and do not take into account practical aspects (measurement errors, component tolerances, etc.). However, they give rigorous limits to the solvability of the fault diagnosis problem and can be considered as an index of efficiency for the used method of fault location. From this point of view, it seems that the results obtained in the above paper could be further improved. In fact, such results allow only to locate a fault between two global ambiguity groups, considering only signed variations of the components (increment or decrement). It is our opinion that the quality of the results obtained by the presented method can be improved exploiting the information given by algorithms for testability and canonical ambiguity group determination. In particular, this information should be taken into account as a preprocessing step in the selection of the ambiguity groups used for neural network training. The measure could be successively considered for a practical verification of the obtained theoretical ambiguity groups. It is worth pointing out that the more recent algorithms for testability computation, based on the symbolic approach [9]‐[11], have a low computational complexity. Hence, the introduction of the testability evaluation phase in the proposed procedure will little increase the computational cost. III. CONCLUSION The method of linear circuit fault diagnosis proposed by Spina and Upadhyaya in this paper presents some interesting aspects as the use of a white noise generator with an artificial neural network. It would be very interesting to further enhance the method by considering the information given by testability and ambiguity groups, achievable with well-known algorithms developed in the last 20 years.