GB2579786A

GB2579786A - Device and method of integrity monitoring and failure predicting for a vehicle system

Info

Publication number: GB2579786A
Application number: GB1820308.3A
Authority: GB
Inventors: Calin Nemes Marian
Original assignee: Continental Teves AG and Co OHG; Continental Automotive Systems SRL
Current assignee: Continental Teves AG and Co OHG; Continental Automotive Systems SRL
Priority date: 2018-12-13
Filing date: 2018-12-13
Publication date: 2020-07-08
Also published as: GB201820308D0

Abstract

An integrity monitoring and failure predicting device comprises: at least one acoustic sensor (2) attached to a monitored component (1) and which is able to convert vibrations generated into acoustic signatures. A digital signal processor (3) pre-processes by filtering, amplifying and digitizing the input signals. A failure mode predictor (4) includes a baseline model unit (41) in the form of a memory that stores a matrix of specific values as a predefined baseline model for each monitored component. A system model unit (42) adapts the predefined baseline model of the monitored system to create system model and an advanced processor (43) reads input signals and creates a current model of the monitored system. A comparator (44) detects deviations between the parameterized baseline model (system model) and the current model and compares the deviation with a threshold. An operational classifier (45) may then classify the deviations in order to predict failure. An associated method is also provided.

Description

Device and method of integrity monitoring and failure predicting for a vehicle system The present invention generally relates to a device and a method of integrity monitoring and failure predicting for a vehicle system, that can be used to detect engine failure, chassis failure, wheel/drive train failures, fan failures, or failure of any part that produces sound/vibration; the predictions are aimed especially at components that can't be monitored electrically in a direct way.

Mechanical devices are prone to failure. The failure rate and a statistically determined mean-time-between-failures are known properties of mechanical elements and assemblies. This leads to servicing of parts at regular intervals for failure prevention. Still, failures may and will happen in the field. These failures cause damage to their own system, to other systems and/or people.

Inmost cases, mechanical on-line failure detections methods are costly and involve special sensors and processing (e.g. strain gauges, torque gauges), therefore they are implemented in highly critical application only (e.g. water dam, aerospace). For example, such a fault prediction system for rotating body using vibration and a method thereof are described by the patent KR10182913431. According to that, vibration data generated by rotation of a rotor are collected, and information entropy is calculated after converting the collected vibration data into a frequency domain so as to measure entropy of the rotor.

However, in the automotive sector, several ways are employed to mitigate mechanical failure development and its effects, such as: mechanical designs robust enough for their application; fallback solutions that protect the driver and passengers inside a vehicle; periodical mandatory verification of vehicle status; exchange of worn-out parts after visual inspection or after a set time/distance. Still, failures can happen even with parts that 5 seem new and functional. As there are no monitoring mechanisms in place, a failure will make itself felt only after it has already happened. An experienced driver will hear and feel the vehicle, subconsciously analysing a known baseline of vehicle vibrations and behaviour to anomalous conditions. In all applicable cases, 10 predicting a failure should be controlled by technical means.

At the design and production stage, the problem is the increased effort and cost to design a system that fails in a controlled way. While this is a good practice, relying only on periodical 15 maintenance is not enough.

In usage, inmost cases the pre-failure behaviour of the vehicle is not influenced enough to cause an average driver to exchange a part that will fail, so there are two options: -the part will fail, leading to down time; - the part will be replaced during regular verification.

In case the failure recovery mechanism works as intended, the main inconvenience for the user is that a failure can be detected only after it happens, causing changes in his/her schedule or increased expenditure, compared to a normal replacement in the service shop. But if the failure also leads to uncontrollable situations, then the importance of an early-warning system is increased as the loss might be more than time and money.

Therefore, the prediction of an emerging failure addresses the following issues: - periodical planned downtime and part replacement, without real need; - unexpected downtime due to an occurred failure; - possible extended damages to the system in which the part has failed; - injury or loss of human lives due to catastrophic failures.

The object of this invention is to obtain a reliable prediction of an emerging failure.

According to the invention, this object is achieved by the subject matters of the independent claims, namely a device for integrity 10 monitoring and failure predicting for a vehicle system, as well as an associated method.

Further advantageous embodiments are the subject matter of the dependent claims.

According to invention, a device of integrity monitoring and failure predicting solves the technical problem by means of: - at least one acoustic sensor attached to each monitored component, said acoustic sensors being able to convert vibrations 20 generated by said monitored components into acoustic signatures; - a digital signal processor supplied with said acoustic signatures provided by said acoustic sensors, and able to filter, amplify and digitize them as input frequency signals; - a failure mode predictor that includes: -a system model unit, namely a data processor that reads vehicle parameters relevant to the monitored mechanical components, and adapts a predefined baseline model of the monitored system to a system model; - an advanced processor that reads said input signals and, 30 based on them, creates a current model of the monitored system; - a comparator able to detect deviations against a set threshold between said system model and said current model of the monitored system; -an operational classifier that classify said detected deviations into failure predictions.

The main advantages arising from a correct failure prediction are: - Preventing damages to the system in which a monitored component has failed; - Prevent injury or loss of human lives due to catastrophic failures, by reacting before the failure occurs.

Also, the technical problem is solved by means of an associated method having the steps of identifying failure modes or suspicious behaviour from acoustic signatures of monitored components, resulting in a prediction that contains information about what kind of failure is developing, which is the confidence level of the prediction and how much time is left until failure is imminent.

The advantages associated to the inventive method are: -reduction of planned downtime and part replacement without real need; - avoidance/reduction of unexpected downtime due to an unexpected failure; - increased user confidence.

Further special features and advantages of the present invention can be taken from the following description of an advantageous embodiment by way of the accompanying drawings, whereby: -Fig. 1 illustrates a first embodiment of a device for integrity monitoring and failure predicting, according to invention, for a system with one monitored component; - Fig. 2 shows another embodiment of the device according to invention, for a system with multiple monitored components; - Fig. 3 shows an acoustic signature as spectrogram, captured for a DC motor supplied by a nominal voltage of 5 V, no load; - Fig. 4 illustrates the acoustic signature of the DC motor from Fig. 3, that presents the behaviour of the DC motor at starting and sudden locking.

Fig. 1 illustrates a first embodiment of a device according to invention, for monitoring a system S consisting into a single component 1, named in the following description "monitored component". More specifically, said monitored component 1 is a DC motor.

In this embodiment, a sensor 2 is attached to monitored component 1, such as to capture low frequency vibrations generated when monitored component 1 is operational (the DC motor is spinning).

More specific, sensors 2 are acoustic-electric sensors (for example, microphones) with frequency response as output signal, appropriate for mechanical resonance and transmission. Those vibrations are formed by juxtaposed oscillations of various frequencies coming from multiple sources, since the monitored system S/component 1 do not operate isolated, but as part of a vehicle that runs on road. It results an acoustic signature, namely a frequency signal varying in time, a waveform difficult to analyse or interpret. Said signal should be band-pass filtered and amplified, digitized and transformed from time-varying amplitude to frequency-in-time (for example, by Fast-Fourier-Transformation, FFT(t)). This transformation reveals the energy level at each frequency of said signal, useful for detecting resonances and structural integrity issues at a glance. While said signals are noisy and hard to analyse in time, the data interpretation is easier in frequency form, especially as nowadays artificial intelligence processing of frequency signals is increasing in popularity. In short, in order to extract a frequency information easier to process, the captured-acoustic signatures are then pre-processed by a digital signal processor 3.

Pre-processed signals are input data for a failure mode predictor 4, along with a series of vehicle parameters (such as vehicle speed, engine rotation per minute etc.), transmitted by means of a communication unit CU of the vehicle. Furthermore, failure mode predictor 4 comprises: - a baseline model unit 41, namely a memory that stores a baseline model (a matrix of specific values) for monitored component 1, 15 in this embodiment said values are specific for the type of motor used; - a system model unit 42, namely a data processor that reads vehicle parameters and adapts said baseline model to a parameterized model of monitored system S, named further on in this description as "system model"; said system model is a mathematical function, that could be written as follows: f ",;" = a- (t,v) + b* f2 (2t,v) + c * fs (3t,v) + d* (4t,v)... Or

f mic = r[Parameter * f n(nt v)] , wherein - frnicis the frequency signal captured by microphone; - a, b, c, are parameters that depend on the motor type; - t, 2t, 3t... are harmonic components of the frequency signal; 30 -v is, in this embodiment, the motor speed; in general, it is an external information needed to be known or measured in order to choose a correct operating point of the model.

Even though such functions are difficult to be defined, it is also 35 possible to train a neural network to "learn" the relations between spectrum components in order to generate such a mathematical model; - an advanced processing unit 43 that reads the pre-processed signals from acoustic sensors 2 and creates a current model of monitored component 1; namely, by this advanced processing the frequency spectrum is separated on frequencies that depend on the source of signal; -a comparator 44 with a set threshold, able to detect deviations between said system model and said current model of monitored 10 component 1; - an operational classifier 45 that classifies said deviations into failure predictions/suspicious behaviour.

Theoretically, the device according to invention uses as input data a processed measured frequency signal of monitored component 1 (DC motor) , and data obtained from a mathematical model by which are defined the relations between the spectrum components of said motor at normal operation. Then, terms like "advanced processing", "baseline model" and "system model" get the following significance: - "advance processing" means, in this context, separating a spectrogram depending on the source of signal; - "baseline model" is a matrix of values depending on the type of motor used; "system model" is the function discussed earlier, that describes the spectral relations, depending on time and motor speed.

Finally, the comparison between current model (what has been 30 measured and processed) and system model means checking the following steps: - Depending on the motor speed, there is a fundamental frequency, with a defined tolerance; if this tolerance is exceeded, then there is an issue; - Depending on the motor speed, all harmonics must have values of frequency and intensity (amplitude) with defined tolerances; if those tolerances are exceeded, there is an issue; - New modes of composing for a signal lead to the detection of a fault; for example: a broken bearing creates new harmonics and variations of their frequencies in time, with a period equal to their rotation speed.

The system model maybe created off-line from recording how the monitored system S vibrates in all normal operation modes, with varying input parameters. When there is a deviation in quality and quantity, below or above the defined tolerances of the system model, said deviation is analysed to identify a possible cause and to predict a failure mode: -If the analysed deviation matches a pattern stored beforehand in a database on a memory (said pattern being defined based on extracted features such as evolution direction and rate of anomalous deviations), then there is a probability to predict a failure mode, a time until failure and a confidence of the prediction; - If the deviation is not described by means of a pattern, the current model still offers indicia regarding anomalous/suspicious behaviour.

Fig. 2 shows another embodiment of the device according to invention, for a system with n monitored components, referenced from 11 to in. Accordingly, baseline model should be as simple or complex as needed for the intended application, taking into consideration a single monitored component or a multitude of monitored components of monitored vehicle system S. It makes sense to consider vehicle speed/road interaction and engine rotation as a source of vibration, superimposed over whatever other mechanical vibration is monitored. All vibrations are a result of the environment and self-operation, all depend on mechanical actuation (e.g. rotation, linear movemenT. noise). Extracted features that express a trend (direction of evolution, rate of anomalous change) are linked to each application and define a deviation. Deviations are therefore specific to application and failure mode, so based on a specific deviation known beforehand and given the application, one may predict the failure mode.

To predict future failure(s), the specific deviation should be 10 analysed qualitatively and quantitatively locally, especially for critical failure cases.

Moreover, it is possible to refine the operation of the described failure mode predictor 4. The sound of the DC motor spinning up and down with no load was recorded and represented by a spectrogram (as baseline model). Fig. 3 shows such an acoustic signature as spectrogram, captured for the DC motor supplied by a nominal voltage of 5 V. In said spectrogram, some frequency variations are more pronounced, allowing their evolution to be tracked. The step of extracting features may be done by employing several Deep Neural Networks and training them to separate vibration patterns and/or their respective sources (namely, which mechanical component generated which specific vibration). In this case, the fundamental frequency band is directly proportional with the rotation speed and the motor supply voltage. The harmonics follow the fundamental, but have an evolution of their own, based on the motor speed. Once the baseline model with no load was defined, it has been used to train the DNN to create a system (parameterized) model. Thus, the motor was connected to a load that can't be driven in random moments, causing the motor to lock, then restart. Fig. 4 illustrates the acoustic signature of the DC motor in case, presenting the behaviour of the DC motor at starting and sudden locking. While this can also be detected through current monitoring, more important is to detect signs that the motor operation is not stable in the frequency events occurring before full lock. Once detected, these signs are recorded as deviation and indicates a future failure.

While certain embodiments of the present invention have been described in detail, those familiar with the art to which this invention relates will recognize various alternative designs and embodiments for practicing the invention as defined by the following claims.

List of references 1 -monitored component -acoustic sensor 2 -digital signal processor 4 -failure mode predictor 41 -baseline model unit 42 -system model unit 43 -advanced processing unit 44 -comparator -operational classifier n -the number of monitored components S -monitored vehicle system CU -vehicle communication unit

Claims

Patent claims 1. Device of integrity monitoring and failure predicting for a vehicle system, said vehicle system having a multitude of 5 monitored components that generate and transmit vibrations, whereby said device comprises: - at least one acoustic sensor attached to each monitored component, said acoustic sensors being able to convert vibrations generated by said monitored components into frequency signals; -a digital signal processor supplied with said frequency signals provided by said acoustic sensors, and able to filter, amplify and digitize them into input signals; - a failure mode predictor that includes: - a baseline model unit, namely a memory that stores a a 15 matrix of specific values as predefined baseline model for each monitored component; - a system model unit, namely a data processor that reads vehicle parameters relevant to the monitored mechanical components, and adapts the predefined baseline model of the 20 monitored system to a system model; - an advanced processor that reads said input signals and, based on them, creates a current model of the monitored system; - a comparator able to detect deviations against a set threshold between said system model and said current model of the 25 monitored system; - an operational classifier that classify said deviations into failure predictions.
2. Method of integrity monitoring and failure predicting for a 30 vehicle system, said vehicle system having a multitude of monitored components that generate and transmit vibrations, characterized by that it comprises the following steos: - acquisition of signals from monitored components belonging to a monitored system and pre-processing them into input signals; - creating a current model of the monitored system based on said input signals; - reading vehicle parameters relevant to said monitored system; - adapting a predefined baseline model of said monitored system 5 into a system model, based on said read vehicle parameters; - detecting deviations between said system model and said current model of the monitored system by comparing them against a set threshold; - classifying said deviations into failure predictions.
3. Method according to claim 2, whereby said failure predictions contains failure mode, time until failure and confidence of prediction.
4. Method according to claim 2, whereby said deviation classifying is performed by training at least one deep neuronal network.
5. Method according to claim 2, whereby said pre-defined baseline model is created by employing and training several deep neuronal networks to separate vibration patterns and/or their respective sources, namely which monitored component generated which specific vibration.