In the automotive industry, more and more assistance systems are being introduced to support the driver. A very well-known component of these assistance systems is the ultrasonic sensor, which is used for parking assistance or automatic parking steering systems. However, these sensors are safety-relevant, therefore as few errors as possible shall occur in the field. In order to detect faulty sensors at an early stage in production a classification based on experience values of faulty sensors shall be performed. This classification is based on machine learning techniques. To cover the entire process of model creation, in this thesis a data mining process has been performed. The necessary foundations have been laid and various techniques that are typically used in data mining were introduced. Then the basics of the sensor functionality and data acquisition in manufacturing was explained. With these basics, the data mining process can then be carried out step by step. First, the business objectives were defined. These include the reduction of defective sensors that are delivered to the customer. In addition, the type of error should be identified for these faulty sensors in order to determine their origin in production and possibly be able to eliminate them. After that, some preprocessing steps could be applied to the data and the first models was able to be trained. Since the optimisation and selection of these models is very complex, automated machine learning techniques were used. These techniques allow the automatic creation of a well-fitting model. This model enables the detection of faulty sensors and their characteristics. As a result, customer satisfaction can be increased by reducing the number of customer returns and the risk of defective sensors in the vehicle. Furthermore, by identifying the type of defect, problems can be detected and corrected in production. This can prevent the production of further defective sensors.