20/08/2010 by Kvaser

Finding faults before they happen with the Leaf Light

CAN interfaces provide the bridge between CAN networks and SATE’s advanced predictive diagnostic solutions.

Fault finding in modern cars is a highly precise science, if the work of Systems & Advanced Technologies Engineering S.r.l. (SATE) is anything to go by. SATE, an Italian company, specialises in the simulation and fault diagnosis of machinery and plant, with automotive systems being a key market. One of SATE’s key customers is a well-known Italian luxury car manufacturer, which has used SATE’s software and consultancy services to diagnose incipient breakdowns in car prototypes during endurance testing.

SATE creates a simulated model of a system, using data from the CAN bus network to create algorithms that monitor the components and predict faults, as well as the impact of wear and tear. The simulations use either ‘transparent box’ models that base their analysis on the physical laws governing a system’s interactions, or ‘black box’ models. The latter are algorithm-based, created using neural networks that are ‘trained’ on a set of real world input and output signals.

Whichever type of simulation model is involved, highly-accurate signal logging from the vehicle’s CANbus network is needed in order to develop and then implement the model e.g. during the training phase in the case of black-box models or parameter tuning for transparent box models. During algorithm training or tuning, signals are logged on the system under normal conditions. Any mismatch between the model’s output and the real-world corresponding quantity implies a fault or an evolving anomaly, such as engine lubrication issues, problems within the cooling system, alternator, clutch or gearbox.

An important benefit of using an algorithmic approach for faultfinding is that there is no need for additional sensors on the CANbus, which can be a source of potential failures, aside from those already present. So in the case of a vehicle, its already-interconnected ECUs will be able to provide enough information about the components and subsystems to give a picture of the reliability and lifetime of the whole car.

‘Hidden’ information from the sensor network

A vehicle has three main sources of sensor-based information: vehicle kinematics (speed, acceleration), engine operation (rpm, water temperature), and driver control actions (steering wheel angle, brake, accelerator pedal position). From these parameters (i.e. without adding more sensors), information such as tyre pressure and temperature can be estimated using SATE’s models. Among the conditions this method is capable of detecting are sensorless tyre deflation, driver behaviour and anomalous driving pattern detection, gearshift classification and synchroniser diagnostics. SATE’s algorithms have also been used to accurately predict small leakages or control anomalies in the engine coolant system, where early detection can prevent potentially severe damage to the motor. Another example is the detection of insufficient oil pressure, whilst it was still within the regular range. In the latter case, SATE provided the customer with a warning of this as early as 5000 to 11000km before engine break down, and well before a test driver could detect it.

SATE uses Kvaser’s Leaf Light CAN to USB interface to connect to the vehicle CANbus, design on-board diagnostic systems and deploy prototype demonstration applications, such as the smart fuel consumption monitoring application it has developed for an HP iPaq for use on trucks. The Leaf Light provides time-accurate and loss free transmission and reception of standard and extended CAN messages, as well as easy connection between any CANbus network and commercial devices equipped with USB ports, such as PDAs, Ultramobile PCs or desktop PCs.

Incipient fault vs threshold-based signal monitoring

At present, the most-commonly employed strategy for fault location by modern vehicle manufacturers is threshold-based signal monitoring, whereby faults are detected when signals exceed a set of thresholds. However, this approach fails to detect incipient faults, which are usually tolerable in the early stages of their development, but which will cause a deterioration of the system performance over time. SATE’s model-based strategy effectively sets dynamic residual thresholds, resulting in faults being detected earlier and averting the false alarms that are often associated with a ‘threshold-based’ method, where excessively narrow or low thresholds have been set.

With so many low cost mobile computing options now available, from Ultramobile PCs to iPhones, SATE’s advanced predictive diagnostic solutions – connected via Kvaser’s CAN to USB or CAN to wireless interfaces – have the potential to be applied to a much wider range of end applications than previously. Where once this kind of dynamic systems modelling was restricted to research and system prototyping applications, it can now be applied to road-going cars, trucks, machinery and plant, to provide early-warning information for fleet managers and maintenance teams. This type of information is also proving beneficial to OEMs that are responsible for providing long-term warranties or full life support of their equipment. And with CAN network technology found in so many applications beyond the automotive sector, it is no surprise to hear that SATE is applying its simulation expertise to fields as diverse as marine and underwater systems, energy generation, and oil and gas.

Image: SATE Fuel Consumption Monitoring and Remote Analysis, for CANROP platforms and vehicle fleets management systems.