This paper explores the potential of installing low-cost multi-sensor units on railway vehicles combined with Machine Learning (ML) and a front-end application to help support maintenance staff in small-to-middle-size industrial networks. Condition information gathered by in-service shunters combined with augmented reality will enhance visual track inspection. Machine learning methods are highly dependent on the quantity and quality of the labeled data. Labeling the data is an expensive task. We propose an innovative approach that integrates the findings of multiple data sources, including cameras and axle box acceleration (ABA) sensors, when the labeled data is scarce or unavailable. In our use case, positioning algorithms enable track-selective data mapping of the collected sensor data. Primarily, camera data allows for switch detection and classification, weather conditions estimation, and detection of dirt, coal, or any anomalies on the rail lines by deploying three different deep-learning-based pipelines. In addition, track irregularities can be detected by the mounted cameras. Axle box acceleration data detect switches, track defects, and heavy dirt on the tracks. The findings based on the camera data are evaluated using the results based on the ABA data and visualized within a front-end application. It ensures that the detections complement each other on track level, enhancing the overall reliability and accuracy of the assessments.