by Andrey Kostyukov, Dynamics Scientific Production Center
The reliability theory has been developing almost 70 years, and all the time people attempt to find out an instance when equipment should fail to be able to prevent its breakdown. Since middle of 50’s last century, a lot of instruments have been developed: vibration pens, portable analyzers, protection and condition monitoring systems. Notwithstanding, even if all these instruments are in use, the probability of failure haven’t been reducing dramatically because of several reasons. Nowadays, some AI systems promise to solve this problem, however, their developers meet either poor and insufficient data or wrong data which is almost or completely not related with a lifespan of certain piece of equipment. As result, statistics based algorithms don’t work relevant to assignment, and their assessment of the probability of failure is seldomly more than 50-60%. So, to be able to provide accurate diagnosis of machinery health, the algorithms of AI should be developed on physics based rules of degradation. Scientific researches are first, and AI is second. In this connection, there are three topics will be considered in this report: nature of errors of degradation process recognition, system with physics based AI, cases of facility’s operation under real-time diagnostics of machinery’s lifespan.