Signal-based diagnostics of the gasoline injection engine

The paper addresses one of the methods of diagnosing the gasoline injection engine available on the vehicle, namely signal-based diagnostics. The most used algorithms for signal-based diagnostics are highlighted. The possibilities offered on an experimental basis are presented and the tests performed in this respect are presented. Time-frequency analysis techniques are applied to detect defects caused in the case of the Audi A6 car engine.


Introduction
In the case of signal-based diagnosis, it is necessary to compare the measured quantity in the event of a fault with the corresponding quantity in the case of normal operation; thus detecting the deviation of the target variable from a normal engine operating mode. For this purpose, the following are used for quantitative or qualitative assessments: first-order statistical characteristics (mean value, variance, standard deviation, entropy, and so on), monospectral (Fourier transform) or bispectral (higher order cumulants), frequency analysis, time-frequency analysis (spectrogram, Cohen class transforms, Stockwell transform, wavelet transform, and so on), cyclostationary analysis (frequency-frequency), probabilistic methods (Bayesian techniques / Bayesian decisions), detection of sudden / abrupt variations etc. As can be seen, these methods do not require mathematical models of gasoline engine operation (as in model-based diagnostics), but involve the existence of constructive elements that provide values of those variables and this aspect shows the increased role of sensors and actuators in engine diagnostics [2,3,6,7]. As noted, signal-based analysis applies to time domain, frequency domain, time-frequency domain, and frequency-frequency domain. The paper will only use time analysis and time-frequency analysis to diagnose the engine.

Experimental research on engine operation without faults
As mentioned, for signal-based diagnosis, functional variables must be compared in the event of a fault and in the faultfree situation. In this sense, experimental research was carried out with an Audi A6 3.0 TFSI Quattro car equipped with a gasoline injection engine ( fig.1) and functional variables were recorded using the Ross-Tech VCDS tester and specialized software for the VAG group. During the experiments, the functional variables were measured for 30 faultfree tests, denoted A1-A30. With the help of these tests, mathematical models of engine operation without faults were established, which allowed the comparison with the cases in which faults were caused.
With n engine speed and  throttle position, all 3 variables being measured.   Similarly, other static characteristics can be established when the engine is running without faults, which are used for comparison with the existence of faults.

Engine operation in the presence of several faults
In this case, some faults were caused during the experiments; for a fault, 30 experimental tests marked D1-D30 were performed.
In this respect, the definition of a fault according to the SAFEPROCESS technical committee should be recalled. Thus, an impermissible deviation of at least one characteristic property / variable of the system from the acceptable / usual / standard / nominal behavior is called a fault; as can be seen from this definition, a fault means a deviation from the nominal value of a parameter or a certain functional variable.
In addition, the error is the quantitative measure of a fault and constitutes a deviation of the system parameters from their nominal values, or a deviation of a variable from its usual value (corresponding to a normal operation).
Finally, a failure means a fault that involves the permanent interruption of the system's ability to perform a required function under specified operating conditions; the failure can therefore be considered as a total fault (100%).  where h(t) represents the analysis window, h * its conjugate, is the frequency, t and u time.
It should also be noted that the spectrogram represents the square of the amplitude of the short-time Fourier transform:   (4) Where t,  and s are time, and  the standard deviation of the variable y.
In the graphs in fig.9c and 10c, on the ordinate axis was represented the normalized frequency, or the relative frequency  r (Nyquist frequency being 0.5 Hz); the time t is rendered on the abscissa axis. The absolute frequency is obtained by multiplying the number on the ordinate axis in fig.9c and 10c with the sampling frequency  s according to relation (1); the axis of the ordinates in fig.9c and fig.10c is the same as the axis of the abscissas in fig.9b and fig.10b, where the graph is arranged rotated to the left by 90 0 .   fig.12c and fig.13c highlight different pictures in zone A. For the timefrequency analysis, the scalogram was used, which represents the square of the amplitude of the wavelet transform [4] and which is defined by the expression (with the scale parameter a): The time-frequency analysis can also benefit from a spatial representation, as it is found in the spectrograms from fig.14, in fig.14b being noticed the decrease of the spectrogram amplitude.

Conclusion
The diagnosis of gasoline injection engines currently uses new methods and algorithms in addition to the classic onboard diagnostic solution. Among these methods is signal-based diagnostics, which benefit from built-in sensors and actuators and new signal processing algorithms. When signal-based diagnostics, time-frequency analysis techniques are used a lot, which specify the time when there is a fault and its size.

Disclosure of conflict of interest
No conflict of interest to declare.