Tool wear has a substantial impression on machining precision, operational stability, and production efficiency in mechanical manufacturing. Even though tool wear status detection and forecasting have been the subject of lots of research, the majority of current methods lack a single, real-time diagnostic capacity that combines both prediction and instantaneous recognition. To address this limitation, a novel deep learning (DL)-based framework is proposed for real-time diagnosis and prediction of tool wear, utilizing a customized architecture named Dynamic Gravitational-tuned Gate Adjusted Long Short-Term Memory (DG-GA-LSTM). Experimental datasets were acquired from a controlled Computer numerical control (CNC) milling setup, capturing multi-sensor data such as cutting force, vibration, and acoustic emissions under varying operational conditions. The acquired signals underwent noise reduction and normalization using Robust Locally Weighted Regression (RLWR) to ensure consistent input quality.The proposed model was implemented in Python and achieved high performance across various evaluation metrics, including MAE (0.0068 mm), RMSE (0.0094 mm), R² (0.9967), Precision (94.68%), and Recall (90.47%) when compared with other existing approaches. This model demonstrates the feasibility and effectiveness of implementing real-time, intelligent diagnostics and prediction for tool wear, contributing to the advancement of predictive maintenance in modern manufacturing environments.