Artificial intelligence and diagnosis

In today's world, systems are becoming increasingly complex. With the development of new technologies and their use in new products, design and maintenance functions are generating tasks that require ever higher levels of qualification. This is not confined to technology alone, but also applies to other fields such as medicine.

Originally, diagnosis was a purely medical concept, referring to the activity of identifying a disease through its symptoms. Of course, this activity forms part of a more global process, whose objective does not end with the identification of illnesses, but above all includes the definition of the care required to cure or relieve the patient. After all, you don't go to your doctor to find out the name of your illness, but to be treated!

This medical vision was then taken up by the technical and industrial world. This time, the aim is no longer to find the causes of a patient's illness, but rather the causes of a failure or breakdown of a physical device. Conceptually, there is no fundamental difference between medical and technical diagnostics. Both involve seeking out the causes of a malfunction in a physical or living system, with a view to repairing or curing it.

In the technical field, the "maintenance" function covers two main categories of activity: activities relating to the management and organization of maintenance, and activities relating to its technical aspects. The latter class is often subsumed under supervision, particularly when dealing with complex systems such as nuclear power plants or electronic rail traffic management systems, for example. It can be broken down into three main tasks: prevention, diagnosis and troubleshooting.

The aim of this article is to present diagnosis from the point of view of artificial intelligence (AI), a field at the frontier of various scientific disciplines such as computer science, automatic control, psychology and so on. It is divided into three sections. The first provides an overview of various definitions. The second presents the main diagnostic methods based on artificial intelligence, focusing on the most representative approaches. Finally, the third section summarizes these approaches, highlighting their properties.