Signal processing method for multihead scales

A signal processing method for processing a weight signal of a weighing device, comprising the following steps:

• a) Repeatedly sampling a weight signal (W) between a first constant value (WK1) and a second constant value (WK2) with a sampling interval (Δt), wherein i discrete values (W(i)) are obtained;
• b) Preprocessing of the weight signal (W), determination and output of resulting characteristics (RK(i)),
• c) Providing at least two, at most i resulting characteristics (RK(i)) to respective inputs (E(p)) of at least one artificial neural network (KNN(1)) as input values, processing the input values (RK(i)) in at least one inner layer (H(m,n)),
• d) Outputting of output values (A(q)) from an output layer (O(q)), wherein an output value (A(1)) is an estimated value (SW) for the second constant value (WK2),
• wherein input layers (I(i)), the at least one inner layer (H(i)) and output layers (O(i)) each comprise neurons which are connected via weighted connections (w(m,n)) to the neurons of the same or respective preceding and/or succeeding layers;
• d1) checking whether the prediction accuracy (PG) is within a predetermined value range and a predetermined standstill criterion (SSK) has been reached;
• and
• d2) when the prediction accuracy (PG) is within a predetermined value range, accepting the estimated value (SW) and when the prediction accuracy is not within the predetermined value range, processing further weight values until a valid value is reached, or adjusting the stability criterion (SSK) and repeat steps a) to d1);
• wherein i artificial neural networks (KNN(i)) are used in series, where i is greater than 1,
• comprising the following further steps:
• e) Reading in the input values (RK(i)) as well as output values (A(i−1, q)) of the respective previous artificial neural network (KNN (i−1)) into a further artificial neural network (KNN(i)), wherein one output value (A(i−1, 1)) is the estimated value (SW) for the second constant value (WK2), and a further output value (A(i−1,2)) is the prediction accuracy (PG) of the respective previous artificial neural network (KNN (i−1));
• f) Calculating output values (A(i,q)) of the current artificial neural network (KNN (i)), wherein one output value (A(i,1)) is the estimated value (SW1) for the second constant value (WK2), and another output value (A(i,2)) is the prediction accuracy (PG1) of the current artificial neural network (KNN (i));
• g) (i−1)-fold repetition of steps e) and f),
• Outputting of the final output values A(q, i), where at least a final estimated value (SW(i)) and a final prediction accuracy (PG(i)) are output.