Abstract
Incipient fault diagnosis of a process is prerequisite to maintain the process safety. But the diagnosis job is very difficult because the symptom of fault in the incipient stage is quite slight. The job becomes more difficult when the incipient fault with multiple causes occurs. It is also not easy to obtain the fault knowledge, i.e., the measurement data in the faulty condition with multiple causes.
This paper describes how to calculate measurement data in a faulty condition with multiple causes from the measurement data each in faulty condition with single cause. Further in this paper we present a macro-architecture of neural networks that can effectively learn and store the data as the knowledge in variety of faulty conditions including faults with single cause and those with multiple causes and that can accurately diagnoses these faults.
The macro-architecture of networks has the hierarchical structure in which the first stage network learns and stores the data of faults with single cause and the networks in the second stage store the data of faults with multiple causes. The allocation of the fault knowledge into networks of the macroarchitecture makes the diagnosing space narrower and leads to efficient and accurate diagnosis even for the faults with multiple causes.
Diagnosis via the macro-architecture of networks for faults with variety of combinations of causes was accurate.
