Due to their affordability and versatility, peristaltic pumps are found in a lot of industrial equipment. All pumps need to be replaced regularly due to tubing degradation. Machine learning algorithms could optimize maintenance interventions on these devices. To develop these algorithms that could perform predictive maintenance classifications on pumps in a non-invasive way, it is mandatory to have properly labeled data. We have recorded, labeled, and formatted a novel dataset that includes lectures from 6 different sensors (3 different accelerometers, 1 gyroscope, 1 magnetometer, and 1 microphone) on running peristaltic pumps in good and bad condition. This work describes the methods used to record and process the data set. All lectures were taken at the same time, which will allow researchers to test sensor fusion techniques. The dataset was made public and accessible using FAIR data principles.