Method and device for training a sensor model for change-point detection

1 . A method for training a data-based sensor model for determining a change-point timing in at least one sensor signal of a cyclic technical process executed by a technical system, the method comprising:

providing a sensor signal time series of a sampled measured value and a change-point timing;

determining an evaluation signal time series by defining an evaluation time window over the sensor signal time series;

determining training datasets having the evaluation signal time series and the change-point timing; and

training the data-based sensor model with the determined training datasets for determining the change-point timing in the at least one sensor signal of the cyclic technical process executed by the technical system,

wherein the change-point timing is provided as a classification vector including a plurality of elements, each element of the plurality of elements assigned to a time point of a plurality of time points within the evaluation time window,

wherein when the change-point timing occurs within the evaluation time window, the classification vector encodes the change-point timing as a value of a corresponding element of the classification vector that corresponds to or comes closest to the time point of the change-point timing,

wherein when the change-point timing occurs before the evaluation time window, the classification vector encodes the change-point timing as a value of a first predetermined element of the classification vector,

wherein when the change-point timing occurs after the evaluation time window, the classification vector encodes the change-point timing as a value of a second predetermined element of the classification vector, and

wherein the data-based sensor model is configured as a deep neural network and is trained using a back-propagation-based training method.

2 . The method according to claim 1 , wherein:

the change-point timing is encoded as the value of the corresponding element of the classification vector, and

remaining elements of the classification vector are encoded with a value different from the value of the corresponding element of the classification vector.

3 . A method for determining a change-point timing in a sensor signal time series of a sampled sensor signal of a cyclic technical process executed by a technical system, comprising:

sampling a sensor signal to obtain a sensor signal time series;

determining an evaluation signal time series as a cut-out of the sensor signal time series according to a predetermined evaluation time window, the evaluation signal time series including a plurality of sensor values each having a corresponding time point of a plurality of time points;

determining an output vector using a data-based sensor model, the output vector including a plurality of elements each having a corresponding index value of a plurality of index values and a corresponding numerical value, each index value corresponding to a time point of the plurality of time points;

searching the output vector to identify a change-point index value of the plurality of index values, the change-point index value having a numerical value closest to a predetermined value; and

assigning the change-point index value to the change-point timing to indicate when (i) the change-point timing is within the predetermined evaluation time window, or (ii) the change-point timing is outside the predetermined evaluation time window,

wherein the technical system is an internal combustion engine having a fuel injection valve, and an actuation time for opening and closing the fuel injection valve is the change-point timing, and

wherein the opening and the closing of the fuel injection valve is controlled based on the assigned change-point index value, which indicates the change-point timing relative to the evaluation time window.

4 . The method according to claim 3 , wherein:

a first element has an index value assigned to an indication that the change-point timing is located before the predetermined evaluation time window, and

a second element has an index value assigned to an indication that the change-point timing is located after the predetermined evaluation time window.

5 . The method according to claim 3 , wherein, after determining that the change-point timing is outside the predetermined evaluation time window, shifting the predetermined evaluation time window relative to the sensor signal time series.