Investigation on the Role of Artificial Intelligence in Measurement System

When transmitters measure process variables like temperature, pressure, flow, level, and quality in process industries, due to wear and tear, ambient conditions, process variations and electromagnetic interference, measured values always deviate from their actual value, and controlled variables will start to oscillate near setpoint. This review paper investigates artificial intelligence impacts on measurement system to improve characteristics of transmitters like linearity, accuracy, sensitivity, resolution, repeatability. Soft computation using artificial neural networks and deep learning for linearization, compensation and error reduction, machine learning for estimation levaraging different algorithms like levenberg marquardt, scaled conjugate gradient, bayesian regularization, are assessed for training, testing and validation in real time and simulation. Hardware approach with soft computation has reduced non linearity error by 84.63% for thermocouple linearization, meanwhile novel hybrid approach using genetic algorithm (GA) and particle swarm optimization (PSO) combined with back propagation neural network (BPNN) have reduced mean absolute percentage error to 1.2 % for industrial weir than conventional hardware approaches using sensors and signal conditioning circuits but at higher computational cost. Challenges like integration to distributed control system via programmable logic controller, huge amount of training data for estimation, real time implementation in measurement systems, overfitting, underfitting, implementation cost are analyzed. Application of artificial intelligence methods and hybrid approaches in measurement systems can drastically improve the operation, maintenance, cost, safety of plant and personnel in real time, where artificial intelligence is still in its nascent stage in instrumentation and control.