INTELLIGENT EMBEDDED SYSTEM FOR DIAGNOSTIC ASSISTANCE IN INSPECTIONS BY PULSED PARASITIC CURRENTS
Pulsed Eddy Current; Embedded Measurement System; Industry 4.0; Neural Network; Extreme Learning Machine.
The use of emerging technologies through the so-called industry 4.0 has provided greater productivity and efficiency. The intelligent production process involves people, machines and equipment, logistic systems and suppliers through a real-time communication and control system that connects them. The continuous monitoring of industrial activities stands out in the optimization of the life cycle, in work safety, assisting to contract industrial insurance, allowing for greater flexibility, precision, reliability, predictability and adaptability to the means of production. In some industrial sectors, equipment and devices are often subject to deterioration by corrosion, which compromises their mechanical and structural property. In extreme conditions, it can cause plant operation failures, human risks and high maintenance costs. One of the challenges in detecting corrosion occurs in equipment with coated metallic material or with thermal insulation. The Pulsed Eddy Current (PEC) technique is well suited for this kind of testing, as it enables inspection on thermally insulated metal parts without the need for coat removal or stopping system operation. However, the magnetic field interaction inside the evaluated specimen generates PEC signals that are not of simple interpretation. Given the above, a processing system was developed electronic, to generate, acquire and process PEC signals for properly identify corrosion in thermally insulated pipes. The system includes analog (current driver and data acquisition) and digital (signal processing, feature extraction and decision support) sub-systems. The proposed signal processing chain comprises feature extraction using discrete Fourier and Wavelet transforms, information compression by Principal Component Analysis and decision support through intelligent classification techniques, using Multi Layer Perceptron - MLP and Extreme neural networks Learning Machine - ELM. The results obtained from tests on a carbon steel tube used in the petrochemical industry indicate the high efficiency of the proposed method. By combining the neural network and data compression, it was possible to obtain an intelligent portable system for rapid assessment and decision support in the field. The best designed neural discriminator had a total efficiency of 99.7% and an average processing time of approximately 2.4 ms.