Abstract
The paper provides an overview of the current challenges of valve inspection in the industry and how they can be overcome with the help of software and artificial intelligence. Studies show that 5-10% of valves in oil and gas plants are suffering from internal leakage which can lead to economic losses, health and safety issues or potentially to environmental pollution. Acoustic emission testing is a relatively modern but already well-established technology for leak detection in valves, however existing solutions show some limitations. These are among others e.g., their complexity of use, they need trained and experienced personnel, the time required to perform analyses and the use of solely proprietary and closed devices. Interviews with experts performing inspections in the field have shown that they are facing challenges such as using the right measuring points, finding the right duration of measurement, applying the right pressure when placing the sensor and finally transferring data and interpreting the results. Research has shown that interpretation of results depends very much on experts' know-how (Singh, Saleh Al Kazzaz, 2003) and reproducible results have thus been difficult to achieve. This causes the difficulty for companies to use the data for further purposes, such as predictive maintenance.
To address this problem, over 2000 measurements in laboratories and simulated leakages in the field have been collected, taking into account different media/pressures/different valve sizes/nominal diameters and valve types. The experience gained and the data collected were used to build a smart and mobile inspection system which is able to standardize inspection processes for leak detection in valves. It was built on the well-established acoustic emission technology, extended with digital features.
Our digital approach has proven that a standardized process is able to significantly reduce the dependence on expert knowhow. By combining AE signals with state-of-the-art signal processing, algorithms and machine learning models, the vast amount of AE signals can be processed within seconds, providing reliable results even in noisy environments. These developments provide companies a new opportunity for in-house valve inspection, a possibility to fight against shortage of skilled labour and increase the reliability of their assets.
