Via the RAPEX rapid alert system for dangerous non-food products, a recall of biodegradable cutlery was initiated in 2024 due to a risk of injury from fragments breaking during use. The failure of such a safety-critical feature can be due to, for example, a design fault or a manufacturing defect. In the era of Industry 4.0, reliability engineering methods are increasingly being characterised and expanded by the application possibilities of artificial intelligence. Particularly within the product development process, opportunities arise to ensure product reliability for safety- and function-critical components. Especially for safety-critical features, it is essential to prove a consistent process capability, which can be achieved through inline inspection as a full inspection process. In this study, images of wooden forks were captured under constant boundary conditions in a test rig set up in a lab. These images serve as a training dataset for machine learning analysis, showing defects in the manufacturing process that could lead to user injury in application. Each fork image was analyzed for defects using Convolutional Neural Networks (CNNs). For the determination of an appropriate model, a comprehensive parameter study is performed. The approach of optimizing the algorithm results and identifying a reliable and reproducible CNN model is presented in detail. The evaluation of the results shows that the found model can detect the defective forks with an accuracy of 81.16%. Under these conditions, a transfer to real operation at an evaluation speed of 4.3 ms per object is possible to a certain extent.