Telematics collects, transfers, and saves machine data like operating data, diagnostic messages, and spatial positions of industrial vehicles equipped with CAN Bus devices. However, the acquired SPN signals are prone to errors and inconsistencies due to transmission errors, changes in environmental conditions, and unpredictable human activities (e.g., fuel theft). Since in industrial vehicle monitoring and management, it is quite hard to define a priori the exceptions that likely occur in the raw SPN series, the idea is to explore the use of unsupervised outlier detection algorithms directly on the raw data or a neural network-based encoding capturing the most salient temporal correlations.