FAQs on Machine Vision Software

Machine vision software is used to solve image processing tasks with wenglor machine vision products in the field of industrial image processing.

wenglor offers the following machine vision software: 

• wenglor uniVision 3 software
• wenglor Discovery Tool software
• uniVision 2 software
• VisionApp 360 Software
• VisionApp Demo 3D Software
• Support software

New software versions are provided for functional enhancements, performance optimizations and bug fixes.

wenglor uniVision 3 is software for setting up wenglor machine vision products for solving tasks in the field of machine image processing. The development environment enables users to automatically evaluate data (e.g. image evaluation) via graphic user interfaces and the creation of configurations instead of conventional programming. wenglor uniVision 3 thus qualifies as a low-code or no-code platform.

Registered users can download and install the uniVision 3 software free of charge via the product detail page DNNF023.

wenglor uniVision 3 is based on the functionality of uniVision 2, but contains numerous new functions, optimizations and bug fixes. uniVision 2 and 3 also support different devices:

• uniVision 2: weQube B50 smart camera, 2D/3D profile sensors, BB1C5 control unit
• uniVision 3: B60 smart camera, MVC machine vision controller

The wenglor uniVision 3 software is supported by the B60 smart camera and the MVC machine vision controller. It is the standard software for all future wenglor machine vision devices.

wenglor uniVision 3 has a toolbox with numerous software modules that can be added flexibly to the job and linked to each other as required.

Templates are predefined uniVision jobs for a specific task (e.g. code reading) that can be loaded onto the uniVision product. The most important modules are already saved and linked in the templates so that only a few parameters need to be adjusted.

Although no programming knowledge is required to operate wenglor uniVision 3, basic knowledge of machine vision processing and parameterization is required.

wenglor uniVision 3 software requires a PC with Windows 10 or Windows 11. Details can be found in the technical data on the product detail page. 

Software and firmware updates for uniVision 3 are published several times a year in order to expand the functional scope of uniVision devices and to continuously improve their stability and performance.

Yes, provided that the devices are compatible with uniVision 3, new software and firmware updates can be downloaded quickly and easily to the corresponding device via the device website. More detailed descriptions of the update process can be found in the operating instructions of the respective device.

Most modules available in wenglor uniVision 3 software can be used several times within a job and can be combined as required. Exceptions are the “Spreadsheet” and “Image Deep OCR” modules (for the B60 smart camera), as well as the interfaces, which can only be used once per job.

Yes, the software can also be expanded later with separate license packages.

Yes, profiles can be read into uniVision offline via Teach Plus or via simulation mode if they are in ply format. For example, profiles can be saved in the VisionApp Demo 3D and simulated in uniVision 3.

Yes, in Teach Plus mode, data (images or profiles) can be saved, deleted and loaded easily via the Image Container Viewer in uniVision. In simulation mode, data (images or profiles) are read from the PC via an unchangeable folder path. 

The DNNF023 uniVision 3 software contains an offline simulator, which can be used largely free of charge and without licensing. Only if the modules

• Image Code 1D

• Image Code 2D

• Image Deep OCR

• Image Pattern Match

• HALCON Script

are to be used offline is the DNNL022 license required.

The wenglor uniVision 3 software can also be used offline for simulation without a device in two different ways:

• Teach Plus mode
• Simulation mode

The Teach Plus mode can be used, for example, to optimize projects with good and bad images taken with the camera. Quick tests for evaluating the software can also be carried out using the examples stored in the software. The offline simulation mode allows the evaluation of the software with image or profile files recorded with third-party hardware or synthetically generated.

The visualization of a job can be set flexibly and freely. Results can be displayed directly in the image as an overlay, for example. Visualization is web-based and can be used on any device with a browser.
 

wenglor uniVision 3 supports all relevant interfaces to control systems and robots so that uniVision devices can be integrated quickly and easily.

With uniVision 3, a robot connection for welding or for robotic vision can be set up for each process instance. The B60 smart camera can therefore be connected to a robot, while the MVC machine vision controller allows 16 individual robot connections for welding and/or robotic vision.

Yes, uniVision 3 has templates for common joint types that make it easier to set up the job.

In addition to a robot, a MLxL 2D/3D profile sensor,a MVC machine vision controller and the uniVision Robotics license package (included in the variant MVCV001 or can be relicensed via the license package DNNL026) are required for weld seam tracking with uniVision 3.

The camera and robot with Robot Vision are calibrated by hand-eye calibration using a calibration object. The camera can be installed static or be located on the end effector of the robot. Professional, rigid and temperature-resistant calibration objects are available in different sizes. For the calibration routine, various positions must be taught in in which the camera sees the calibration object so that the relationship between the camera and robot is determined.

The uniVision 3 software enables communication with robots from different manufacturers. The open Robot Vision API can also be used to connect to robots whose type is not yet officially supported.

Robot Vision is supported in wenglor uniVision 3 by the B60 smart camera and by the MVC machine vision controller with the machine vision cameras of the BBVK or BBZK series.

Calibration plates are used in measuring applications to clean up optical distortion and ensure precise conversion from pixel to millimeter values. This takes place in the wenglor uniVision 3 software via the Image Calibration module.
In addition, a calibration plate enables simple and fast calibration in robot vision applications. This process also involves coordinate matching, thus eliminating the distortion caused by the optics. For precise calibration, the calibration plate should be completely in the camera’s visual field and cover at least half of the visual field. Paper-printed calibration patterns result in inaccurate calibration. Opaque panels (e.g. ZVZJ001) are suitable for incident light applications, transparent (e.g. ZVZJ005) for transmitted light applications.

If the relationship between the camera and robot does not change, recalibration is not necessary.

In wenglor uniVision 3 software, data from several different objects can also be found with an image capture and sent to the robot in order to optimize the cycle time for pick-and-place tasks. This means that the robot has to move into the detection position less frequently and can directly grasp other objects that have already been found.
 

Any desired offsets in x and y can also be set up in wenglor uniVision 3 software for picking in pick-and-place applications so that the object can be gripped at the tip, for example.

In the Pick and Place application, an individual object height can be set up for each object type so that different object types can be gripped at different heights. 

Various object types can be easily taught in using wenglor uniVision 3 software. The object type can then be sent directly to the robot.

 

Teaching in objects is easiest in wenglor uniVision 3 using the Pattern Match and Locator modules. 

The “Device Robot Vision” module in wenglor uniVision 3 enables direct communication between 2D cameras and robots.

Yes, the running of HALCON scripts can also be tied to a specific device. This prevents a project from simply being copied to other devices using HALCON script.
 

HALCON scripts can be encrypted to prevent unwanted changes to the script.

Data such as taught-in contour models can be stored permanently and independently of the platform in the HALCON dictionary.

There are numerous examples of HALCON scripts that show in a simple format which data types are supported and how applications can be implemented easily.

The typical workflow for working with HALCON scripts is as follows:

• Recording of a Teach+ file with real data
• Creation of the HALCON script with the recorded data in the HDevelop software
• Loading the HALCON Script into wenglor uniVision 3 software in the HALCON Script module

The image data required to create a HALCON script is recorded in a Teach+ file with real data.

The uniVision ecosystem enables flexible data exchange between all uniVision modules and the HALCON Script module. Numerous interfaces (e.g. PROFINET, EtherNet/IP) are available on the uniVision product. The results from the HALCON Script module can thus be output directly and flexibly via the uniVision interfaces. The flexible web-based visualization also enables flexible and individual display of results from the HALCON Script module – even directly in the image!
 

Yes, the HDevEngine is already running on uniVision devices. HALCON scripts can thus be run directly on uniVision devices. This allows the focus to be placed on the application solution (software)!
 

HALCON scripts can also be created with other HALCON versions. However, the compatibility information with the version HALCON 22.11 used on uniVision devices must be observed.
 

uniVision devices run version HALCON 22.11.

The following data types can be transferred from uniVision modules to the HALCON Script module (inputs) or returned from the HALCON Script module to other uniVision modules (outputs):

• Iconic variables
  • Images
  • Regions
  • XLDs
• Control variables
  • Integer
  • Real
  • String

Yes, the standard software modules in wenglor uniVision 3 software can be combined with HALCON scripts as desired. Flexible data exchange between the modules is possible!

wenglor uniVision 3 thus enables a combination of parameterization and programming:

• Parametrization:
  Standard tasks can be easily completed with the standard uniVision modules from the uniVision Toolbox 
• Programming: Complex tasks can be solved in HDevelop with HALCON scripts.

HALCON is MVTec’s extensive program library with image processing algorithms. The development environment for HALCON is the HDevelop software.

Learn more about HALCON MVTec here!
 

HALCON scripts created in MVTec’s HDevelop software can be loaded in the HALCON Script module in the uniVision software and run on the uniVision product (e.g. B60). The HDevEngine required for this is already pre-installed on the uniVision product.

The AI Lab enables the creation of your own AI models directly from the user interface without the need for additional steps outside the AI Lab. AI models created therein are executed in uniVision 3 via the “Image AI” module. The “Image ONNX” module is also a uniVision 3 module for executing AI models and is suitable if your own network architectures are to be used or image data for training purposes must not leave the company network. However, this requires further steps outside of the AI Lab, for example in-house training and export as an ONNX model. For more information, see our Github repository. Both the “Image ONNX” module and the “Image AI” module are included in the “uniVision AI” license package.

After a payment plan expires, the account automatically switches back to the Free plan. The AI Loop is then no longer available, but the user can still access their data records and delete them if necessary. If the Free plan also expires, access to the AI Lab account will no longer be possible. The account is then in a grace period: It can be reactivated with all data by entering a new license key. If this is not done, the account will be permanently deleted after the expiry of the grace period.

wenglor reserves the right to use image data from Free and Free+ accounts to improve its own models. Private data records are excluded.

Yes, the transfer function can be used to transfer an entire data set including images, labels and models to another AI Lab account. The data set is moved (not copied) and is then only available in the target account. If the data set is to remain available in the original account, it can be duplicated before transfer.

Within a plan, internal users can access records together. Credits, memory and other resources are shared. Collaboration with external users who have their own AI lab plan is possible via specifically released data records.

Plans can be extended by stacking – identical licenses add runtime and credits. An ongoing plan can be replaced by another plan at any time. If limits such as memory, users or credits are reached, data that is no longer required can be deleted or you can switch to a higher plan.

Yes, in uniVision 3, rules can be defined according to which images of a specific class or with a score below a defined threshold value are automatically added to a data set in the AI Lab via the AI Loop. This is also possible during operation.

All trained model versions are saved. In uniVision 3, either the last trained model or the primary model can be provided. The latter is defined by the user in the AI Lab. The validation metrics, predictions and heatmaps in the AI Lab always refer to the last model trained.

Trained AI models can only be executed on wenglor hardware. However, the training data can also be created with third-party devices and then uploaded. The AI Lab is designed for PC use and is not optimized for smartphones or tablets.

No permanent internet connection is required. For image upload, weHub can buffer data and transfer it to the AI Lab as soon as a connection is established again. No internet connection is required for inference on the device itself.

There is no fixed limit on the number of classes. It is recommended that the data set covers all real-life image scenarios. It is also important to have as balanced a ratio of images per class as possible. A minimum of 5 images per class is required, and a minimum of 50 images is recommended for reliable results. For more information on AI in image processing, visit our technology page.

As standard, the AI Lab generates quantified INT8 models as they run faster via the B60’s integrated NPU. Non-quantified FP32 models can also be created and run; these run on the CPU (ARM) with the B60 and are therefore slower than a quantified INT8 model on the NPU.

Images can be uploaded in full resolution, but are automatically scaled to the appropriate input size for training and execution. Only the relevant image sections should be uploaded, as otherwise important image information may be lost during scaling – cropping to the relevant area, e.g. in uniVision, is recommended. Depending on the hardware, only executable model sizes can be selected in the AI Lab. The AI model size is derived from the AI input image size and the AI model architecture and directly influences the inference speed.

The upload rate depends on the network, image size and number of devices. Several frames per second per device are common. The following applies to the classification: 1 credit = up to 5,000 images, 2 credits = up to 10,000 images, then 1 credit per 5,000 images. The basic value for object detection is 3 credits (up to 5,000 images); 1 credit is added for each additional 5,000 images.

A training session with ≤ 500 images at 320 px usually takes around 5 minutes. Depending on the workload, the training can initially be queued (training queue), which can extend the waiting time until the start. Results may vary slightly between different training sessions, as random elements increase robustness. Optionally, the user can be notified by e-mail as soon as the training is completed.

Individual images can be selectively excluded from the training data set and assigned to the blind test data set instead. These images are not incorporated into the training or the regular validation, but are only used for an additional, independent (blind test) validation after the final model has been selected.

The division takes place automatically via a K-fold validation. In addition, the user can specifically exclude individual images from the training data set and assign them to a blind test data set, which is only used for additional, independent validation after the final model has been selected.

An existing model cannot be retrained directly – every training is always based on the complete data set. If you want to improve a model with new data, new images can be added to the existing data set and then completely retrained. In practice, this is equivalent to retraining.

The AI Lab supports various image formats (e.g. JPEG, PNG, BMP, TIF) and resolutions that are automatically adjusted. Loss-free compressed formats such as PNG, BMP or TIF should be preferred over JPEG. For best results, all images should be of comparable quality.

Yes, all data is stored in Europe in accordance with GDPR. In payment plans, all rights remain with the user. The cloud storage is BSI C5 certified and data can be deleted yourself after the end of the plan.

All data is encrypted using TLS and AES-256, stored multiple times redundantly and backed up automatically. This protects it against loss and unauthorized access.

Yes, several B60 smart cameras can be uploaded to the same data set in parallel via weHub, as long as the plan allows free “connected devices”.

The connection between the AI Lab and uniVison takes place via weHub, via which images are uploaded to the AI Lab and trained AI models are transferred back to uniVision.

The evaluation report is available for checking the AI model quality. It shows the most important key figures for accuracy and serves as proof, e.g. for a factory acceptance.

weHub is software for detecting and managing wenglor machine vision devices in the network. It enables adaptation of the network configuration, automated upload of images to the AI Lab, and download of AI models from the AI Lab to the wenglor machine vision hardware.

weHub replaces the wenglor Discovery Tool: It offers the same functions as device search and network configuration for wenglor Machine Vision devices, plus the bridge function to connect the cloud-based AI Lab with the offline devices.

weHub is a standard software for all wenglor machine vision devices. The hardware currently supports the B60 smart camera, the MVC machine vision controller and the 3D sensors of the ShapeDrive G4 series.

A PC with the Windows 10 or Windows 11 operating system is required for using weHub. Please refer to the “Technical Data” section on the product details page for further details on the system requirements for operating the software.

Yes, weHub finds all supported devices, even if they are in a different subnet.

weHub indicates the status normal operation, warning or error with details of the warning or error message. Example: The temperature of the machine vision device exceeds a critical value. A warning message with detailed information “Temperature is too high” then appears in the software.

It is often difficult to find the network settings in your PC settings. weHub displays the network settings of the PC directly without having to call up the PC settings.

The device website can be accessed easily via weHub. The IP address of the device is not required.

weHub can be used to assign a name of your choice to each device so that it is easy to distinguish between several devices.