Diagnosis of pumping machine working conditions based on transfer learning and ViT model

Abstract

The dynagraph card is an important tool for diagnosing work conditions in rod pump systems. However, a key problem is that actual dynagraph cards may have different working conditions yet exhibit similar graphs. This reduces the classification performance of existing models and leads to poor diagnostic accuracy. To solve this problem, we propose a novel method based on transfer learning and ViT models for diagnosing the working conditions of rod pumping systems. Specifically, we first pre-trained the ViT model using the ImageNet-1k dataset, and then fine-tuned the weights of the model using the actual dynagraph card dataset. This transfer learning way not only significantly reduce the training time of the model, but also effectively improve the accuracy of the working condition diagnosis. To evaluate the performance of the proposed method, we compare it with ResNet, DenseNet, and RegNet models. Our experimental results demonstrate that our method achieves a work condition diagnosis accuracy of 0.9060, which is higher than other methods by 0.2–0.3. Moreover, our method performs well on the problem of different work conditions but graphically similar dynagraph cards. Therefore, our transfer learning and ViT model-based method can better solve the practical problems of dynagraph cards and meet the needs of oilfield sites.