Diagnostics

(Not available yet in the current release)

In the Diagnostics Window, a selected model of a module network can be validated against a selected data set. On the basis of residual correlation tests, information is provided on the validity of the different modules in the network. The tools can be used for fault detection and diagnosis, by confronting the data from a possibly faulty network, to a model of the healthy network. By combining the results of the different correlation tests, the particular location of a fault can be determined.

Correlation tests that are being performed:

• $R_{\varepsilon_j}(\tau)$, the autocorrelation of the prediction error at node $w_j$;
• $R_{\varepsilon_ju_k}(\tau)$, the crosscorrelation of the prediction error at node $w_j$ and an excitation signal at node $w_k$;
• $R_{v_ju_k}(\tau)$, the crosscorrelation of the simulation error at node $w_j$ and an excitation signal at node $w_k$;

while for the network model use can be made of either

• the full noise model $H(q)$,
• the topology of $H(q)$ only, or
• no information on $H(q)$.

Current implementation restrictions:

• Not available yet for diffusively coupled networks (DCN's)

Reference

• Y. Shi, S.J.M. Fonken and P.M.J. Van den Hof (2024). Fault detection and diagnosis using the dynamic network framework. IFAC PapersOnLine, Vol. 58-15 (2024), pp. 384-389. Proc. 20th IFAC Symposium on System Identification, 17-19 July 2024, Boston, MA, USA.