FB_VN_ReadNeuralNetwork

FB_VN_ReadNeuralNetwork 1:

This FB reads a neural network from a file on the target pc.

Syntax

Definition:

FUNCTION_BLOCK FB_VN_ReadNeuralNetwork
VAR_INPUT
    sFilePath           : STRING;
    ipDestNeuralNetwork : Reference To  ITcVnNeuralNetwork;
    aInputShape         : Reference To  TcVnArray33_UDINT;
    bRead               : BOOL;
    nTimeout            : TIME;
END_VAR
VAR_OUTPUT
    bBusy               : BOOL;
    bError              : BOOL;
    nErrorId            : UDINT;
END_VAR

 Inputs

Name

Type

Default

Description

sFilePath

STRING

Full path of the file or relative path to the default directory on the target pc

ipDestNeuralNetwork

Reference To ITcVnNeuralNetwork

Returns the neural network

aInputShape

Reference To TcVnArray33_UDINT

Optional input shape for testing the execution of the neural network model. The error code INCOMPATIBLE is returned if the execution fails. Setting aInputShape = 0 or aInputShape[0] = 0 disables the testing of the execution.

bRead

BOOL

Reading the file is triggered by a rising edge at this input.

nTimeout

TIME

VISION_ADS_TIMEOUT

Indicates the time before the function is cancelled.

 Outputs

Name

Type

Description

bBusy

BOOL

This output remains TRUE until the function block has executed a command, but at the longest for the duration supplied to the 'nTimeout' input. While bBusy = TRUE, no new command will be accepted at the inputs.

bError

BOOL

This output is switched to TRUE as soon as an error occurs during the execution of a command. The command-specific error code is contained in ‘nErrorId’. If the function block has a timeout error, 'bError' is TRUE and 'nErrorId' is 1861 (hexadecimal 0x745). Is reset to FALSE by the execution of a command at the inputs.

nErrorId

UDINT

Contains the ADS error code or the command-specific error code of the last executed command. Is reset to 0 by the execution of a command at the inputs.

Further information

The function block FB_VN_ReadNeuralNetwork loads a neural network from a file asynchronously into the PLC. The file must be in the standardized exchange format Open Neural Network Exchange (ONNX), as only this format is supported.

FB_VN_ReadNeuralNetwork 2:

Loading modified neural networks

When loading a neural network, the Vision Service caches the data internally. Therefore, it is not possible to load modified neural networks in the standard system. To do this, adjust the Cache settings or see the sample Serial loading of images into the PLC for an alternative.

Parameter

Some parameters are described centrally in the section File Access Parameters.

Loaded neural network (return value)

The loaded ITcVnNeuralNetwork can be reached via the interface pointer ipDestNeuralNetwork after a successful read process and optional test of the execution.

Testing the execution

The optional parameter aInputShape can be used to transfer an input format that is used to test the execution of the neural network before loading it into the PLC. This test is helpful, especially for models with dynamic input formats, to ensure that the model can be successfully executed with the specified format. If the execution is not successful, a HRESULT = ADSERR_DEVICE_INCOMPATIBLE is returned.

The following information can be used to specify an input format for the test, for example. The execution test can also be disabled with aInputShape[0] := 0;.

aInputShape[0] := 4;
aInputShape[1] := 1;
aInputShape[2] := 224;
aInputShape[3] := 224;
aInputShape[4] := 3;

Application

Loading a neural network, without testing the execution (aInputShape := 0;), from the path C:\TcVision\NeuralNetwork.onnx looks like this, for example:

fbReadNeuralNetwork(
    sFilePath           := 'C:\TcVision\NeuralNetwork.onnx', 
    ipDestNeuralNetwork := ipReadNeuralNetwork,
    aInputShape         := 0;
    bRead               := TRUE,
    nTimeout            := T#500MS);

IF NOT fbReadNeuralNetwork.bBusy AND NOT fbReadNeuralNetwork.bError THEN
    // Neural Network is accessible in ipReadNeuralNetwork
END_IF

Required License

TC3 Vision Base

System Requirements

Development environment

Target platform

PLC libraries to include

TwinCAT V3.1.4024.59 or later

PC or CX (x64) with PL50, e.g. Intel 4-core Atom CPU

Tc3_Vision