Error correction - "INTAS-OPEN-97-0606"

Reference Number: INTAS-97-0606

Development of an Intelligent Sensing Instrumentation Structure I.S.I.S.

Using Neural Networks in ISIS

Possible methods of error correction in ISIS

The main goal of the error correction in ISIS is the receiving of corrected value of physical quantity lC. It is necessary to calculate value of the correction factor - (DSl + lSt) and make correction (SUM). The existing ways of correction factor calculation are:

1	To sum value of no corrected physical quantity l NC with value of the correction factor for the measured value of physical quantity l NC at time of measuring. l C = l NC - (D Sl 1 + D St 1). The higher ISIS level changes D Sl 1 and D St 1 at changing of physical quantity l NC and time of exploitation t . The higher ISIS level operates in real time scale and middle level executes only SUM.
2	To sum value of no corrected physical quantity l NC with value of the correction factor for time of measuring according to measured value of physical quantity l NC. l C = l NC - (D Sl + D St 1). The higher ISIS level changes D St 1 during exploitation time and middle level calculates D Sl as function of l NC. The higher ISIS level operates not in real time scale (on low degree) and middle level must operates in real time scale.
3	To sum value of no corrected physical quantity l NC with value of the correction factor according to time of measuring and according to measured value of physical quantity l NC. l C = l NC - (D Sl + D St ). The higher ISIS level changes D Sl and D St if it is necessary (the life of previous correction factor is finished). The higher ISIS level operates not in real time scale (on average degree because it must control time of correction factor life) and middle level must operates in real time scale.
4	To sum value of no corrected physical quantity l NC with value of the correction factor according to time of measuring and according to measured value of physical quantity l NC. l C = l NC - (D Sl + D St ). The higher ISIS level changes D Sl and D St by request from the middle level (the life of previous correction factor is finished). The higher ISIS level operates not in real time scale (on high degree) and middle level must operates in real time scale (because it must control time of correction factor life additional). The middle level must have a timer in this case.

Possible presentation of correction factor at the middle ISIS level

As number (digital code)
As set of numbers (table)
As formula (first degree of polynomial, second degree of polynomial, any other formula)
As weights and thresholds of neural network with constant structure
As set of neural networks with constant structures and their weights and thresholds
As neural network with various structure (by reprogramming) and its weights and thresholds

Possible operations of the neural networks in ISIS

N	Operation	Neural network input	Neural network output
1	Influence Quantity Correction	SIQD - Digital Code of Sensor Signal of Influence Quantity SIQ	Correction Factor c(SIQD)
2	Physical Quantity Calculation	SIC Ö Corrected Digital Code Sl D of Sensor Signal of Physical Quantity	No Corrected Physical Quantity l NC
3	Sensor Sensitivity and Offset Error Correction	No Corrected Physical Quantity l NC (and (if possible) SIQD - Digital Code of Sensor Signal of Influence Quantity SIQ)	Sensitivity and Offset Error D Sl
4	Sensor Drift Correction (1)	Duration of Sensor Exploitation (one point because PNN has one input in this case)	Sensor Drift D St
5	Sensor Drift Correction (2)	Values of Sensor Drift D St for previous time points (calibration points approximated by ANN, number of points defined by number of PNN inputs)	Sensor Drift D St