This study investigates the application of a multi-spectral photodiode sensor technology and deep learning for real-time monitoring of aluminium busbar laser welding processes. Specically focusing on different defect detection, which undermines product quality. Utilizing a single sensor approach with a multi wavelength photodiode, the signal captures features from the welding process, which are subsequently analyzed using a ConvLSTM model. Importantly, the study reveals that the backrejections are a critical source of information for assessing welding quality. Additionally, the light at wavelengths around 400 and 700 nm wavelength provides significant information, likely to the change in excitation state of the aluminium vapour, but also due to the vapourazation of the native aluminium oxcide layer. These features are identified as significant predictors of aluminiumlaser weld defects, highlighting their importance and potential to enhance the predictive accuracy of defect detection. By integrating a machine learning techniques with real-time monitoring, this research demonstrates the potential to improve the precision and reliability of welding operations, especially in the production of critical components for the automotive industry, such as cell-tobusbar joints.