In the context of the use of vehicles for demanding tasks, problems arose, among other things, with regard to the load beyond specified limits, as well as difficulties in recognizing and correcting systematic errors and in introducing needs-based maintenance.
The foregoing resulted in requirements for the development of an autarkic system which can be easily adapted to different vehicles and which, thanks to a consequent structured structure in hardware and software, meets special requirements and which can be used cost-effectively and reliably.
TROUT developed and tested a Vehicle Data Analyzing System, VeDAS for short, on behalf of Rheinmetall Landsysteme GmbH.
It is used for the automatic acquisition of vehicle data and their evaluation. Communication with the associated evaluation system takes place quickly and securely (safety & security) using a mobile storage medium or wireless communication.
The detected data includes the accelerations and vibrations experienced by the vehicle as well as its location. In addition, speed and distance covered are recorded. In order to take possible environmental influences into account, temperature and humidity are determined. In addition, a structure-borne sound microphone provides information about the operating status of the monitored vehicle. In addition, data provided by an engine control unit, for example, can be taken into account via a CAN bus interface.
In addition to vehicle data acquisition, VeDAS has a logbook function for documenting maintenance activities. Maintenance intervals and deadlines are noted here for the vehicle assemblies entered.
Structure of the evaluation software
Via the drop-down menu “Vehicle”, the user selects the desired vehicle for which data has been recorded in advance. The following tabs then contain the vehicle data and the possibility of evaluating them.
- Vehicle data: Creation and modification of vehicle master data, summary of all vehicle data, graphical representation of terrain / road section
- Usage profile: Evaluation of the vehicle data after a selectable period.
- Graphical representation: route of the vehicle, graphical representation of the vehicle data, structure-borne noise, impact diagram
- Assemblies: Organization of the condition-monitored components
- Settings: language setting, directories, export settings, definition of limit values and correction factors.
- VeDAS data transfer: configuration of the VeDAS data stick, data import from the data stick.
- Vehicle-specific identification numbers and identifiers as well as the date for the last maintenance, the next maintenance and the last data import can be found under master data.
In the configuration area, the map material and the route are called up for the “graphic display”.
If a corresponding map is loaded, the GPS data available in the route are transferred to this map. The user can track the course of the vehicle.
The color of the markings on the card indicates the speed or other parameters, depending on the operator selection.
Each marking on the map can be selected by the user in order to receive additional information such as temperature and humidity. In addition, the user can select a time frame around this point in order to switch to the “Graphical evaluation” tab and analyze the area around this point.
“Kategorien anzeigen” enables the user to select and display measurement data in a diagram.
A sequence of values or dates is considered a time series here. When observing different time series, there is a growing interest in what future values or events will look like. In the case of needs-based maintenance, sensors give signals at certain time intervals, which are processed using specially developed methods so that statements can then be made about future developments. A reliable forecast about the future is usually not possible, but if different methods are used, temporal developments can be forecast with some accuracy, as is the case here.
Our approach of analyzing multidimensional time series is based on deep neural networks (deep learning) in order to be able to adequately map the complex technical system of the vehicle.
Neural networks are able to create non-linear relationships between inputs and outputs without prior knowledge, which is why they have proven themselves as a method in this area as well. In particular, we use the multilayer perceptron (MLP) here, it is a multilayer feedforward network (ffn).
The selected method of condition monitoring now specifies the maintenance intervals and ensures the availability of the vehicle. In a further expansion stage, it will also be possible to determine the time for the next engine oil change. Complex algorithms can be used to draw conclusions about the aging process of the oil used from the sensed data.