Enhancing the safety of energy pipeline transportation: A mechanism-morphology confidence evaluation method

As an efficient way to transport gaseous or liquid energy such as hydrogen, natural gas, and ammonia, energy pipelines have been widely developed and expanded. With its usage time increases and the transported components become increasingly complex, accurate state assessment methods are particularly important to improve the overall efficiency and reliability of energy transmission networks. Internal corrosion failure is a major factor affecting the safety of gas pipeline transportation, thus accurate and rapid identification of its failure types is crucial for shortening the cycle of failure evaluation, reducing the cost of failure analysis, and enabling the swift adoption of countermeasures. Presently, the number of sampled pipeline corrosion images is limited, and there are numerous interfering factors. Traditional methods find it difficult to accurately extract key corrosion features. This paper proposes a mechanism-morphology confidence evaluation method for internal corrosion failure of pipelines. The mechanism-morphology confidence evaluation functions were constructed for five types of corrosion. Gradient-weighted class activation mapping was used to determine the model's area of focus towards the image and to calculate the confidence of the identification results. The physical prior information was used to guide model training so as to allow the model to learn image features while ensuring the consistency of corrosion mechanism. Finally, verify the recognition effect in combination with the on-site pipelines. The results show that the improved model with an overall identification accuracy of up to 93.7%, demonstrating that the proposed method can significantly enhance the classification accuracy, thereby improving the model's credibility and interpretability.