Issues with equation count in NF

[Francesco Casella]

Please check the ​FilterWithDifferentiation model with the NF. When checking the model, I get

Modelica.Blocks.Examples.FilterWithDifferentiation has
15 equation(s) and 15 variable(s).

and I can also count 15 equations if I flatten the model. However, when compiling the model the following error is obtained:

Error: Too many equations, over-determined system.
The model has 13 equation(s) and 11 variable(s).
Error: Internal error Transformation Module 
PFPlusExt index Reduction Method Pantelides failed!

As there are some equations involving complex functions, I suspect there is something wrong in how the DAE structure is built in that case.

[Willi Braun]

This seems to be an issue in removeSimpleEquation, since the model is balanced when we enter the Backend. Output dumpdaelow:

Variables (15)
========================================
1: der3.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real 
2: der3.u:VARIABLE(flow=false )  "Connector of Real input signal" type: Real 
3: der2.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real 
4: der2.u:VARIABLE(flow=false )  "Connector of Real input signal" type: Real 
5: der1.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real 
6: der1.u:VARIABLE(flow=false )  "Connector of Real input signal" type: Real 
7: Bessel.uu[3]:VARIABLE(flow=false protected = true )  type: Real  [3]
8: Bessel.uu[2]:VARIABLE(flow=false protected = true )  type: Real  [3]
9: Bessel.uu[1]:VARIABLE(flow=false protected = true )  type: Real  [3]
10: Bessel.x[3]:VARIABLE(flow=false )  "Filter states" type: Real  [3]
11: Bessel.x[2]:VARIABLE(flow=false )  "Filter states" type: Real  [3]
12: Bessel.x[1]:VARIABLE(flow=false )  "Filter states" type: Real  [3]
13: Bessel.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real 
14: Bessel.u:VARIABLE(flow=false )  "Connector of Real input signal" type: Real 
15: step.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real 


Equations (15, 15)
========================================
1/1 (1): step.y = Bessel.u   [dynamic]
2/2 (1): Bessel.y = der1.u   [dynamic]
3/3 (1): der1.y = der2.u   [dynamic]
4/4 (1): der2.y = der3.u   [dynamic]
5/5 (1): step.y = step.offset + (if time < step.startTime then 0.0 else step.height)   [dynamic]
6/6 (1): Bessel.uu[1] = Bessel.u / Bessel.u_nominal   [dynamic]
7/7 (1): der(Bessel.x[1]) = Bessel.r[1] * (Bessel.x[1] - Bessel.uu[1])   [dynamic]
8/8 (1): der(Bessel.x[Bessel.nr + 2 - 1]) = Bessel.a[1] * Bessel.x[Bessel.nr + 2 - 1] + Bessel.ku[1] * Bessel.uu[Bessel.nr + 1] -
 Bessel.b[1] * Bessel.x[Bessel.nr + 2]   [dynamic]
9/9 (1): der(Bessel.x[Bessel.nr + 2]) = Bessel.b[1] * Bessel.x[Bessel.nr + 2 - 1] + Bessel.a[1] * Bessel.x[Bessel.nr + 2]   [dyna
mic]
10/10 (1): Bessel.uu[2] = Bessel.x[1]   [dynamic]
11/11 (1): Bessel.uu[Bessel.nr + 1 + 1] = Bessel.x[Bessel.nr + 2]   [dynamic]
12/12 (1): Bessel.y = Bessel.gain * Bessel.u_nominal * Bessel.uu[Bessel.nr + Bessel.na + 1]   [dynamic]
13/13 (1): der1.y = der(der1.u)   [dynamic]
14/14 (1): der2.y = der(der2.u)   [dynamic]
15/15 (1): der3.y = der(der3.u)   [dynamic]

After removeSimpleEquations: optdaedump

########################################
pre-optimization module removeSimpleEquations (simulation)
########################################


unspecified partition
========================================

Variables (11)
========================================
1: der3.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real
2: der3.u:STATE(1,der3.y)(flow=false )  "Connector of Real input signal" type: Real
3: der2.u:STATE(1,der3.u)(flow=false )  "Connector of Real input signal" type: Real
4: der1.u:STATE(1,der2.u)(flow=false )  "Connector of Real input signal" type: Real
5: Bessel.uu[3]:VARIABLE(flow=false protected = true )  type: Real  [3]
6: Bessel.uu[2]:VARIABLE(flow=false protected = true )  type: Real  [3]
7: Bessel.uu[1]:VARIABLE(flow=false protected = true )  type: Real  [3]
8: Bessel.x[3]:STATE(1)(flow=false )  "Filter states" type: Real  [3]
9: Bessel.x[2]:STATE(1)(flow=false )  "Filter states" type: Real  [3]
10: Bessel.x[1]:STATE(1)(flow=false )  "Filter states" type: Real  [3]
11: step.y:VARIABLE(flow=false )  "Connector of Real output signal" type: Real


Equations (13, 13)
========================================
1/1 (1): der3.y = der(der3.u)   [dynamic]
2/2 (1): der3.u = der(der2.u)   [dynamic]
3/3 (1): der2.u = der(der1.u)   [dynamic]
4/4 (1): der1.u = Bessel.gain * Bessel.u_nominal * Bessel.uu[Bessel.nr + Bessel.na + 1]   [dynamic]
5/5 (1): der(Bessel.x[Bessel.nr + 2]) = Bessel.b[1] * Bessel.x[Bessel.nr + 2 - 1] + Bessel.a[1] * Bessel.x[Bessel.nr + 2]   [dynamic]
6/6 (1): der(Bessel.x[Bessel.nr + 2 - 1]) = Bessel.a[1] * Bessel.x[Bessel.nr + 2 - 1] + Bessel.ku[1] * Bessel.uu[Bessel.nr + 1] - Bessel.b[1] * Bessel.x[Bessel.nr + 2]   [dynamic]
7/7 (1): der(Bessel.x[1]) = Bessel.r[1] * (Bessel.x[1] - Bessel.uu[1])   [dynamic]
8/8 (1): Bessel.uu[1] = step.y / Bessel.u_nominal   [dynamic]
9/9 (1): step.y = step.offset + (if time < step.startTime then 0.0 else step.height)   [dynamic]
10/10 (1): Bessel.uu[1] = Bessel.x[1]   [dynamic]
11/11 (1): Bessel.uu[2] = Bessel.x[1]   [dynamic]
12/12 (1): Bessel.x[2] = Bessel.x[1]   [dynamic]
13/13 (1): Bessel.uu[3] = Bessel.x[3]   [dynamic]

[Willi Braun]

When we pass removeSimpleEquations, which can be achieved with --removeSimpleEquations=new | --removeSimpleEquations=none the model fails in the index reduction with some differentiation issue for the equation:

Bessel.y = Bessel.gain * Bessel.u_nominal * Bessel.uu[Bessel.nr + Bessel.na + 1]

this equation looks in the old instantiation:

Bessel.y = Bessel.gain * Bessel.u_nominal * Bessel.uu[3]

The final parameters are also replaced in the old Frontend.

I guess the old removeSimpleEquation module acts also wrong, because it assumes all ASUB expressions have been replaced.

[Francesco Casella]

I guess the back-end is currently built on the assumption that all structural parameters are evaluated, right?

Should the NF do that, or should the BE be extended to take care of that?

[Francesco Casella]

The model now simulates and verifies correctly.