Run AI Models Directly on wenglor Hardware with ONNX!
ONNX (Open Neural Network Exchange) is an open standard that makes it possible to integrate AI models into a wide range of applications across different platforms. ONNX converts AI models created in different development environments, such as PyTorch or TensorFlow, into a universal format. As a result, these models can then be used on other platforms.
User-friendly and efficient
Save development time and resources!
The “Image ONNX” module reduces the work needed to prepare data, as AI models can be easily imported and run on uniVision 3.
Flexibility in use
Combine AI and rule-based image processing!
The “Image ONNX” module can be integrated into the uniVision 3 workflow as well as combined with rule-based image processing modules as necessary.
Scalability and platform independence
Use AI models on different hardware!
ONNX models can be used on both the B60 series smart cameras and on machine vision controllers.
Easy integration
Reap the benefits of using existing industrial interfaces!
Results can be processed directly by the existing uniVision 3 communication infrastructure in your production process.
Using AI Models with ONNX on uniVision 3
The “Image ONNX” module provides AI functions for image classification. Existing ONNX models are seamlessly integrated into your workflow and can be used directly for industrial image processing. The heatmap visualization can be used to check how well models work.
Superior Performance with High Hardware Compatibility
With the uniVision 3 software, the wenglor machine vision hardware provides the perfect basis for running ONNX models efficiently.
- The B60 series smart cameras are equipped with an integrated neural processing unit (NPU) that is specially optimized for accelerating AI calculations. This means that demanding tasks like image classification can be performed quickly and reliably.
- The MVC series machine vision controllers use the OpenVINO hardware acceleration platform to run ONNX models efficiently while also minimizing resource consumption.
Seamless Integration, Maximum Productivity: Workflow with ONNX and uniVision 3
Creation of AI Model in ONNX Format
Conversion
Integration into uniVision 3
What Is AI Classification in Image Processing?
AI classification in image processing uses convolutional neural networks (CNNs). These neural networks analyze images and assign them to a predefined class based on certain characteristics. The system then learns to distinguish typical properties of the respective classes.
If a new image is taken, it is given a probability distribution (score) across all possible classes by the AI model. The class with the highest probability is considered to be the model’s prediction. This enables, for example, automatic distinction between fault-free (OK) and faulty (NOK) components. In addition to a binary classification (OK/NOK), the model can also distinguish between multiple error classes.
In order to make the decision-making logic of the AI more transparent, a heatmap is also generated that visualizes the image areas that have had the greatest influence on the model during classification. This allows users to better understand which characteristics led to the decision given by the model.
One example is quality control in injection molding production. Various types of errors can occur here, which the AI model accurately detects and classifies. As these errors often vary in shape and form, AI classification offers greater flexibility and accuracy than rule-based methods. This system is made especially powerful by a combination of AI and classic image processing, which enables robust and reliable quality controls.
If a new image is taken, it is given a probability distribution (score) across all possible classes by the AI model. The class with the highest probability is considered to be the model’s prediction. This enables, for example, automatic distinction between fault-free (OK) and faulty (NOK) components. In addition to a binary classification (OK/NOK), the model can also distinguish between multiple error classes.
In order to make the decision-making logic of the AI more transparent, a heatmap is also generated that visualizes the image areas that have had the greatest influence on the model during classification. This allows users to better understand which characteristics led to the decision given by the model.
One example is quality control in injection molding production. Various types of errors can occur here, which the AI model accurately detects and classifies. As these errors often vary in shape and form, AI classification offers greater flexibility and accuracy than rule-based methods. This system is made especially powerful by a combination of AI and classic image processing, which enables robust and reliable quality controls.
Example of fault-free component (OK)
Example of faulty component (NOK)
Example of faulty component (NOK)
Example of faulty component (NOK)
Licensing
The “Image ONNX” module is part of the “uniVision AI” license package and is available for all products with the wenglor uniVision 3 software.
The following licensing rules apply:
The following licensing rules apply:
- B60 series: DNNL031 - uniVision AI B60 license
- MVC series: DNNL032 - uniVision AI MVC license
- uniVision Offline Simulator: This module can be used without an additional license.
