Practical aspects of applying artificial intelligence in metrology

Abstract

The article considers artificial intelligence as one of the main elements of digital transformation and promising directions of its application in metrology. The main attention is paid to the use of artificial neural networks as part of measuring instruments and measuring systems for obtaining measurement results in cases where the measurement function is unknown, insufficiently defined or too complex for algorithmic formalization. It is noted that in practice problems with a partially uncertain function, when, in addition to the deterministic basis, there is an additional unknown component that has a significant impact on the measurement result, often arise. A simulation experiment was carried out to solve such a measurement problem using a neural network model. In the experiment, a measurement function with a linear deterministic basis and an additional nonlinear component, which is approximately 10% of the relative standard deviation and is a priori unknown (according to the conditions of the problem), was used. The experimental results confirmed the practical possibility and high efficiency of using artificial neural networks to solve such measurement problems. The neural network model, under conditions of a noisy training dataset corresponding to real measurement conditions, almost completely restored the measurement function, despite the fact that the neural network model used was linear, and the additional component of the measurement function was nonlinear. In this experiment, due to the use of a neural network, the accuracy of measurements was improved by approximately an order of magnitude. Access to the machine code that implements this simulation experiment is provided.