This work presents an Artificial Intelligence (AI) system designed to provide real-time predictions for defect formation in additive layer manufacturing of composites. The system relies on a set of in-line laser sensors that continuously monitor the Automated Fibre Placement (AFP) manufacturing process for composites. These sensors feed data into an AI system, which is capable of learning both the spatial and temporal characteristics of the material as it is being deposited. This innovative AI system is intended to not only project the current state of the deposited material during manufacturing, but also to identify and classify manufacturing defects. Ultimately, it aims to forecast the manufacturing status in the future. This capability enables real-time prognosis of defect formation. Consequently, the control system responsible for controlling the manufacturing process will have the capacity to predict the occurrence of defects before they manifest. This proactive ability provides an opportunity to intervene in the manufacturing process, thereby mitigating the formation of defects before they become problematic. The proposed AI system is trained in multiple stages. Initially, an autoencoder is trained to extract spatial feature from images, generated by laser profilometry, encoding these features into a latent vector. Next, a deep LSTM network is trained to project the future status of the production process in latent space. Finally, a classifier based on deep neural networks is trained to classify the encoded future projections into non-defective or 3 categories of commonly occurring defects in AFP manufacturing. The system has been successfully demonstrated on a lab scale AFP production machine.