Expert systems vs. human expert
Expert systems have in recent years become an integral package of many online monitoring systems and diagnostic test equipment sold to transformer manufacturers, service providers and end users. They are also commonly used in oil laboratories for providing standardized reports with recommendations on dissolved gas analysis and oil quality tests results. This article discusses briefly the limitations of algorithms typically used in expert systems and in more detail their shortcomings. The article also provides examples of why expert systems have not and cannot replace the final review and assessment of the test results by a human expert considering that reports produced, automatically, by such expert systems will not cover the complex facts when important and costly decisions have to be made.
Keywords: expert system, human expert, limit values, reliability, credibility, intuition
In most of the commercially available monitoring systems, it is now the current state of art for online or offline measuring systems to be supplied along with integrated expert systems. The big advantage is that results based on the established analysis algorithms and limit values are now available and can be provided in an auto-printed standard format report, in which case no further investigation may be necessary. That way, the user may believe that he has obtained an accurate condition analysis of his transformers. With the expert now inside the instrument, the user may wonder whether he still needs an external human expert at all.
It remains to be seen to what extent such expert systems can replace a specialist who will not only have many years of experience, but also an intuition based on learning on how to go beyond standard responses in order to look for and find the true facts behind the data, and evaluate the total complex situation and the consequences.
The reliability of the various expert systems that are available remains quite controversial. What should the user do after receiving conflicting results on identical data from different expert systems? Should the view of the majority prevail, or is there a better approach?
In the following, relevant examples will be used to demonstrate that expert systems are only able to deliver standard answers, which eventually may mislead the user. A “flesh-and-blood” expert is absolutely indispensable in order to obtain true understanding of the facts and receive suggestions of the right measures to both understand the problem and correctly address the situation.
Transformers are costly and complex assets which require competent management in order to avoid costly and unexpected failures. It will be a long time before we see a truly reliable expert system working independently from human experts, possibly based on highly sophisticated fuzzy logic or similar systems.
Let’s not forget: transformers are hand-crafted by humans, making incalculable human behaviour inherent in the system at all times.