Ultrasonic non-destructive evaluation is of extreme importance in the oil and gas industry, especially for assets and structures subjected to conditions that accelerate failure mechanisms. Despite being widely spread, ultrasonic non-destructive methods depend on a specialized workforce, thus being errorprone and time-consuming. In this context, pattern recognition methods, like machine learning, fit conveniently to solve the challenges of the task. Hence, this work aims at applying artificial intelligence techniques to address the interpretation of data acquired through ultrasonic non-destructive evaluation in the context of the oil and gas industry. For that purpose, this dissertation involves three case studies. Firstly, ultrasonic guided wave signals are used to classify defects present in welded thermoplastic composite joints. Results have shown that, when using features extracted with autoregressive models, the accuracy of the machine learning model improves by at least 72.5 percent. Secondly, ultrasonic image data is used to construct an automatic weld diagnostic system. The proposed framework resulted in a lightweight model capable of performing classification with over 99 percent accuracy. Finally, simulation data was used to create a deep learning model for estimating the severity of corrosion-like defects in pipelines. R2 results superior to 0.99 were achieved.