This work proposes an end-to-end integrated model to automatically detect run-time errors in CNC milling automation composed of temporal convolutional networks (TCN), attention-based long short-term memory (LSTM) networks, and neuro-fuzzy systems. The approach applies the greatest features of the individual sub-parts in intelligent manufacturing systems to support high accuracy, minimal downtime, and improved quality of the produced products. The primary objectives are to reduce false negatives and positives, enhance the quality of products, and raise the precision in detecting CNC milling automation defects. Through reducing loss of time and real-time detection of defects, the model has been constructed with the aim to enhance efficiency. Methods: The approach of the model proposes merging TCN, attention, and neuro-fuzzy systems. TCN holds the dependencies which arise over a specified time series of data, attention-LSTM assists in locating important time steps, and neuro-fuzzy systems operate in a random environment. Robust fault detection is enabled by the utilization of these blocks. Results: The integrated system performed significantly better than current methods in accuracy, precision, recall, and F1-score. It showed higher defect identification rates and reduced false positives and negatives by a large margin. Overall, this solution combination improves CNC milling quality and reduces downtime. The experiments show the effectiveness of the nLSTM model for a broader spectrum of job loads on smart manufacturing systems.