V0.11.0

Project.toml

@@ -8,6 +8,7 @@ DiffEqCallbacks = "459566f4-90b8-5000-8ac3-15dfb0a30def"
 DifferentiableEigen = "73a20539-4e65-4dcb-a56d-dc20f210a01b"
 DifferentialEquations = "0c46a032-eb83-5123-abaf-570d42b7fbaa"
 FMIImport = "9fcbc62e-52a0-44e9-a616-1359a0008194"
+FMISensitivity = "3e748fe5-cd7f-4615-8419-3159287187d2"
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
 Optim = "429524aa-4258-5aef-a3af-852621145aeb"
 Printf = "de0858da-6303-5e67-8744-51eddeeeb8d7"
@@ -18,13 +19,13 @@ ThreadPools = "b189fb0b-2eb5-4ed4-bc0c-d34c51242431"
 
 [compat]
 Colors = "0.12.8"
-DiffEqCallbacks = "2.26.0"
 DifferentiableEigen = "0.2.0"
-DifferentialEquations = "7.8.0"
-FMIImport = "0.15.8"
-Flux = "0.13, 0.14"
+DifferentialEquations = "7.10.0 - 7.11.0"
+FMIImport = "0.16.2"
+FMISensitivity = "0.1.0"
+Flux = "0.14"
 Optim = "1.7.0"
-ProgressMeter = "1.7.0"
+ProgressMeter = "1.7.0 - 1.9.0"
 Requires = "1.3.0"
 ThreadPools = "2.1.1"
 julia = "1.6"

README.md

@@ -41,6 +41,7 @@ You can evaluate FMUs inside of your loss function.
 - building and training NeuralFMUs consisiting of multiple FMUs
 - building and training FMUINNs (PINNs)
 - different AD-frameworks: ForwardDiff.jl (CI-tested), ReverseDiff.jl (CI-tested, default setting), FiniteDiff.jl (not CI-tested) and Zygote.jl (not CI-tested)
+- use `Flux.jl` optimisers as well as the ones from `Optim.jl`
 - ...
 
 ## What is under development in FMIFlux.jl?
@@ -62,6 +63,7 @@ To keep dependencies nice and clean, the original package [*FMI.jl*](https://git
 - [*FMIImport.jl*](https://github.com/ThummeTo/FMIImport.jl): Importing FMUs into Julia
 - [*FMIExport.jl*](https://github.com/ThummeTo/FMIExport.jl): Exporting stand-alone FMUs from Julia Code
 - [*FMICore.jl*](https://github.com/ThummeTo/FMICore.jl): C-code wrapper for the FMI-standard
+- [*FMISensitivity.jl*](https://github.com/ThummeTo/FMISensitivity.jl): Static and dynamic sensitivities over FMUs
 - [*FMIBuild.jl*](https://github.com/ThummeTo/FMIBuild.jl): Compiler/Compilation dependencies for FMIExport.jl
 - [*FMIFlux.jl*](https://github.com/ThummeTo/FMIFlux.jl): Machine Learning with FMUs (differentiation over FMUs)
 - [*FMIZoo.jl*](https://github.com/ThummeTo/FMIZoo.jl): A collection of testing and example FMUs

examples/src/advanced_hybrid_ME.ipynb

@@ -65,10 +65,9 @@
     "| 1.  | Enter Package Manager via         | ]                         |\n",
     "| 2.  | Install FMI via                   | add FMI                   | \n",
     "| 3.  | Install FMIFlux via               | add FMIFlux               | \n",
-    "| 4.  | Install FMIZoo via                | add FMIZoo                | \n",
-    "| 5.  | Install DifferentialEquations via | add DifferentialEquations | \n",
-    "| 6.  | Install Plots via                 | add Plots                 | \n",
-    "| 7.  | Install Random via                | add Random                | "
+    "| 4.  | Install FMIZoo via                | add FMIZoo                |  \n",
+    "| 5.  | Install Plots via                 | add Plots                 | \n",
+    "| 6.  | Install Random via                | add Random                | "
    ]
   },
   {
@@ -100,7 +99,7 @@
     "using FMIFlux\n",
     "using FMIFlux.Flux\n",
     "using FMIZoo\n",
-    "using DifferentialEquations: Tsit5\n",
+    "using FMI.DifferentialEquations: Tsit5\n",
     "using Statistics: mean, std\n",
     "import Plots\n",
     "\n",
@@ -120,7 +119,7 @@
     "\n",
     "![svg](https://github.com/thummeto/FMIFlux.jl/blob/main/docs/src/examples/img/SpringPendulum1D.svg?raw=true)\n",
     "\n",
-    "In contrast, the model *SpringFrictionPendulum1D* (*realFMU*) is somewhat more accurate, because it includes a friction component. \n",
+    "In contrast, the model *SpringFrictionPendulum1D* (*fricFMU*) is somewhat more accurate, because it includes a friction component. \n",
     "\n",
     "![svg](https://github.com/thummeto/FMIFlux.jl/blob/main/docs/src/examples/img/SpringFrictionPendulum1D.svg?raw=true)"
    ]
@@ -156,9 +155,9 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### RealFMU\n",
+    "### *fricFMU*\n",
     "\n",
-    "In the next lines of code the FMU of the *realFMU* model from *FMIZoo.jl* is loaded and the information about the FMU is shown."
+    "In the next lines of code the FMU of the *fricFMU* model from *FMIZoo.jl* is loaded and the information about the FMU is shown."
    ]
   },
   {
@@ -175,15 +174,15 @@
    },
    "outputs": [],
    "source": [
-    "realFMU = fmiLoad(\"SpringFrictionPendulum1D\", \"Dymola\", \"2022x\")\n",
-    "fmiInfo(realFMU)"
+    "fricFMU = fmiLoad(\"SpringFrictionPendulum1D\", \"Dymola\", \"2022x\")\n",
+    "fmiInfo(fricFMU)"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "In the function fmiSimulate() the *realFMU* is simulated, still specifying the start and end time, the parameters and which variables are recorded. After the simulation is finished the result of the *realFMU* can be plotted. This plot also serves as a reference for the other model (*simpleFMU*)."
+    "In the function fmiSimulate() the *fricFMU* is simulated, still specifying the start and end time, the parameters and which variables are recorded. After the simulation is finished the result of the *fricFMU* can be plotted. This plot also serves as a reference for the other model (*simpleFMU*)."
    ]
   },
   {
@@ -201,15 +200,15 @@
    "outputs": [],
    "source": [
     "vrs = [\"mass.s\", \"mass.v\", \"mass.a\", \"mass.f\"]\n",
-    "realSimData = fmiSimulate(realFMU, (tStart, tStop); recordValues=vrs, saveat=tSave)\n",
+    "realSimData = fmiSimulate(fricFMU, (tStart, tStop); recordValues=vrs, saveat=tSave)\n",
     "fmiPlot(realSimData)"
    ]
   },
   {
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "The data from the simulation of the *realFMU*, are divided into position and velocity data. These data will be needed later. "
+    "The data from the simulation of the *fricFMU*, are divided into position and velocity data. These data will be needed later. "
    ]
   },
   {
@@ -273,7 +272,7 @@
    },
    "outputs": [],
    "source": [
-    "fmiUnload(realFMU)"
+    "fmiUnload(fricFMU)"
    ]
   },
   {
@@ -282,7 +281,7 @@
    "source": [
     "### SimpleFMU\n",
     "\n",
-    "The following lines load, simulate and plot the *simpleFMU* just like the *realFMU*. The differences between both systems can be clearly seen from the plots. In the plot for the *realFMU* it can be seen that the oscillation continues to decrease due to the effect of the friction. If you simulate long enough, the oscillation would come to a standstill in a certain time. The oscillation in the *simpleFMU* behaves differently, since the friction was not taken into account here. The oscillation in this model would continue to infinity with the same oscillation amplitude. From this observation the desire of an improvement of this model arises.     "
+    "The following lines load, simulate and plot the *simpleFMU* just like the *fricFMU*. The differences between both systems can be clearly seen from the plots. In the plot for the *fricFMU* it can be seen that the oscillation continues to decrease due to the effect of the friction. If you simulate long enough, the oscillation would come to a standstill in a certain time. The oscillation in the *simpleFMU* behaves differently, since the friction was not taken into account here. The oscillation in this model would continue to infinity with the same oscillation amplitude. From this observation the desire of an improvement of this model arises.     "
    ]
   },
   {
@@ -343,7 +342,7 @@
     "\n",
     "In order to train our model, a loss function must be implemented. The solver of the NeuralFMU can calculate the gradient of the loss function. The gradient descent is needed to adjust the weights in the neural network so that the sum of the error is reduced and the model becomes more accurate.\n",
     "\n",
-    "The loss function in this implementation consists of the mean squared error (mse) from the real position of the *realFMU* simulation (posReal) and the position data of the network (posNet).\n",
+    "The loss function in this implementation consists of the mean squared error (mse) from the real position of the *fricFMU* simulation (posReal) and the position data of the network (posNet).\n",
     "$$ e_{mse} = \\frac{1}{n} \\sum\\limits_{i=0}^n (posReal[i] - posNet[i])^2 $$\n",
     "A growing horizon is applied, whereby the horizon only goes over the first five values. For this horizon the mse is calculated."
    ]
@@ -382,7 +381,7 @@
    "source": [
     "#### Function for plotting\n",
     "\n",
-    "In this section the function for plotting is defined. The function `plotResults()` creates a new figure object. In dieses figure objekt werden dann die aktuellsten Ergebnisse von *realFMU*, *simpleFMU* und *neuralFMU* gegenübergestellt. \n",
+    "In this section the function for plotting is defined. The function `plotResults()` creates a new figure object. In dieses figure objekt werden dann die aktuellsten Ergebnisse von *fricFMU*, *simpleFMU* und *neuralFMU* gegenübergestellt. \n",
     "\n",
     "To output the loss in certain time intervals, a callback is implemented as a function in the following. Here a counter is incremented, every twentieth pass the loss function is called and the average error is printed out."
    ]
@@ -413,7 +412,7 @@
     "                     legendfontsize=8, legend=:topright)\n",
     "    \n",
     "    Plots.plot!(fig, tSave, posSimple, label=\"SimpleFMU\", linewidth=2)\n",
-    "    Plots.plot!(fig, tSave, posReal, label=\"RealFMU\", linewidth=2)\n",
+    "    Plots.plot!(fig, tSave, posReal, label=\"fricFMU\", linewidth=2)\n",
     "    Plots.plot!(fig, time, posNeural, label=\"NeuralFMU\", linewidth=2)\n",
     "    fig\n",
     "end"
@@ -658,7 +657,7 @@
    "source": [
     "#### Comparison of the plots\n",
     "\n",
-    "Here three plots are compared with each other and only the position of the mass is considered. The first plot represents the *simpleFMU*, the second represents the *realFMU* (reference) and the third plot represents the result after training the NeuralFMU. "
+    "Here three plots are compared with each other and only the position of the mass is considered. The first plot represents the *simpleFMU*, the second represents the *fricFMU* (reference) and the third plot represents the result after training the NeuralFMU. "
    ]
   },
   {
@@ -716,7 +715,7 @@
    "source": [
     "### Summary\n",
     "\n",
-    "Based on the plots, it can be seen that the NeuralFMU is able to adapt the friction model of the *realFMU*. After 1000 training steps, the curves already overlap quite well, but this can be further improved by longer training or a better initialization."
+    "Based on the plots, it can be seen that the NeuralFMU is able to adapt the friction model of the *fricFMU*. After 1000 training steps, the curves already overlap quite well, but this can be further improved by longer training or a better initialization."
    ]
   },
   {

src/FMIFlux.jl

@@ -71,6 +71,10 @@ using FMIImport: fmi2SetTime, fmi2CompletedIntegratorStep, fmi2GetEventIndicator
 using FMIImport: fmi2SampleJacobian, fmi2GetDirectionalDerivative, fmi2GetJacobian, fmi2GetJacobian!
 using FMIImport: fmi2True, fmi2False
 
+import FMIImport.FMICore: fmi2ValueReferenceFormat
+
+include("optimiser.jl")
+include("hotfixes.jl")
 include("convert.jl")
 include("flux_overload.jl")
 include("neural.jl")
@@ -80,7 +84,7 @@ include("deprecated.jl")
 include("batch.jl")
 include("losses.jl")
 include("scheduler.jl")
-#include("optimiser.jl")
+include("compatibility_check.jl")
 
 # from Plots.jl 
 # No export here, Plots.plot is extended if available.