Is It Really Possible To Predict Breakdowns Or Derates?
Yes! There are a couple of different approaches that fleets can use to predict breakdowns on the road. These approaches can also predict when a vehicle will derate. Derating is the operation of a vehicle at less than its rated power. Fleet managers want to avoid derating as the vehicle is essentially useless and must be brought into the maintenance shop for repairs.
One approach taken by fleets is to use sensor and fault code data. Another approach is to use prognostic software that leverages machine learning algorithms. It is important to understand their nuances before deploying them on your fleet.
Sensor and Fault Code Data
Most fleets are now equipped with telematics systems (hardware and software). This allows for fleets to monitor their vehicles remotely and in real-time. A telematics system can monitor specific sensors and detect fault codes. In general, sensor data is used to monitor a vehicle’s mechanical health. A few examples of what sensors measure include engine speed, tire pressure, battery performance, exhaust oxygen, intake pipe pressure, and vehicle acceleration. Setting thresholds and alerts on specific sensors can be used to predict breakdowns. However, this requires analyzing historical data and performing experiments (trial and error) to determine the sensor thresholds. Faulty sensors or newer vehicles with less or no historical data make this approach ineffective.
Vehicle and component manufacturers leverage sensor data with proprietary algorithms to generate fault or DTC (diagnostic trouble) codes. Some fault codes are made visible on the driver’s dashboard e.g. check engine light. The lights on the dashboard help inform the driver or fleet manager when a vehicle should be taken in for maintenance. They also help the technician understand where to start looking when doing their diagnostics. This allows the driver to act and bring the vehicle into the maintenance shop for service. The fault codes that are not visible on the dashboard can be used by auto mechanics to diagnose issues. Reporting and alerting on specific fault codes can be effective at predicting breakdowns. But, it does require in-depth knowledge of the fault codes and vehicle’s condition. Another issue is that a fault code is typically generated after progressive damage occurs, resulting in higher repair costs. There are also many cases where sensor fault codes do not help – the fault code is trigger when the vehicle derates or breaks down.
Predictive maintenance (prognostic) software can understand the condition a vehicle has been used in, its current state, and then predict the future allowing you to know when vehicles in your fleet are due for maintenance before a breakdown on the road occurs. Identifying issues before faults are generated requires the use of advanced machine learning algorithms. These algorithms are trained to detect anomalies from multiple sensors at the same time.
In general, an algorithm is as good as the data it was trained on. So, algorithms that are trained using large and diverse data sets (e.g. many vehicle makes and models) are very effective at detecting issues before fault codes and downstream damage occurs.
Let’s go over an example of progressive damage. In diesel vehicles, fuel is injected in the exhaust stream as part of the regeneration process to react with the DOC (Diesel Oxidation Catalyst) to burn off accumulated soot in the DPF (Diesel Particulate Filter). Suppose that the fuel injector is leaky. This would cause unintended elevated temperatures (e.g. above 1000F) in the exhaust for an extended period. Initially, this would result in a damaged DPF, which can cost about $8,000 in parts to repair. Without immediate repair and maintenance, the sustained high temperatures could result in damage to the SCR as well. Replacement costs of an SCR can be about $15,000 in parts.
Since the regeneration process requires exhaust gas temperature to increase, this problem cannot be detected by just monitoring the exhaust gas temperature. If using your telematics system to monitor this vehicle, you would detect this issue after the DPF is damaged when a fault (e.g. ‘Soot level 2 error’) is logged. With prognostics software using advanced prediction algorithms, the issue will be detected earlier, avoiding SCR and DPF damages. In this example, the saving associated with avoiding progressive damage would be about $23,000, not including labor.
Preteckt is the leading Prognostics-as-a-Service platform today trusted by fleets across North America. Having worked with large fleets, Preteckt has large amounts of vehicle and failure data. This data helped create very accurate prediction algorithms that cover many vehicle makes and models.