A Statistical Approach for Leak Monitoring of Highly Noisy Metallic Pipelines

A method for leak monitoring of metallic pipelines operating in the highly noisy environment of an oil refinery is proposed in this work, that can be utilized in industrial applications of similar nature. This method is based on evaluating a set of features extracted from the acquired acoustic signal of a pipeline. It relies on defining thresholds that are constantly updated based on previously acquired feature values. The algorithm behind the operation runs in short segments and produces an output based on the current feature values and their active thresholds. An occurring leak event can introduce a noticeable change in these values and cause them to exceed their specified limits. The decision about the presence of a leak and, eventually, whether a leak is detectable or not, is determined by how many and which of the features are out of the threshold limits at a given time. Some additional steps were introduced to account for random transient events that are present during noise measurements and can cause false alarms. The algorithm is associated with a set of parameters, that can be adjusted in such a way that the system can adapt to the noise characteristics of the pipeline under inspection. Also, the algorithm can begin operating after acquiring only a reference noise measurement during its initialization and without the absolute need for a training phase with artificial leaks. The proposed algorithm’s overall performance was evaluated with on-field measurements that took place on multiple operating pipelines in the facilities of an oil refinery, as well as a laboratory setting.