Machine vision supports manual assembly process

There, machine vision does not replace employees, but instead works "hand in hand" with them. Due to the very high variance of components, full automation is not possible. Nevertheless, the integrated machine vision software MVTec MERLIC significantly enhances quality assurance in the assembly process by displaying all relevant information for assembling the component to the production worker, then verifying the process and documenting quality.

Machine vision - as the "eye" of production - performs a variety of inspection tasks, reads text and codes, and controls robot movements, among other functions. The advantage is clear: automation makes processes more efficient, and quality increases because machine vision operates more accurately and faster than any human could. However, there are applications in which machine vision cannot (yet) fully leverage its specific advantages. This is the case when the variance of components is very high and production volumes are very low. In such situations, humans clearly are at an advantage.

Nevertheless, machine vision can still provide support - for example, as part of a digital assistance system for manual assembly steps. Endress+Hauser has developed such an application.

Endress+Hauser Flow, headquartered in Reinach, Switzerland, is a global leader in flow measurement technology and fluid management solutions. "At Endress+Hauser, our goal is to be technologically leading, to leverage innovations, and to actively drive them forward. This applies not only to our products, but also to our assembly processes," explains Julius Krause from the Industrial Engineering department at Endress+Hauser.

In his role, he supports global production sites in the design, procurement, and commissioning of camera-based inspection systems and digital assistance systems. Krause was also instrumental in developing the digital assistance system for assembly processes. Here, machine vision software MVTec MERLIC was called into action. MERLIC is integrated into Endress+Hauser's proprietary web-based software, which functions as a Manufacturing Execution System (MES).

Among other things, the MES checks whether the device is at the correct process step, whether the workstation is suitable for that step, and whether the logged-in user is authorised to perform the process. Once approval is granted, the assembly process begins. The MES displays work instructions - for example, which materials are to be used - in both image and text form.

In the background, MERLIC is launched with the recipe data and parameters relevant to the process step. The machine vision software confirms that the step has been executed correctly, thereby providing additional assurance of product quality. Once the assembly process is completed, this is reported back to the MES. The results, process data, as well as image data are recorded.

Collaborating to ensure maximum product quality

Before the process was converted to machine vision, quality assurance was typically carried out through double checks by two employees, along with additional tests. "The goals of further development were to relieve production staff, provide optimal support, and ensure the highest level of product quality. In addition, in-process inspection allows errors to be corrected easily and prevents failures in subsequent process steps or end-of-line tests," explains Julius Krause, outlining the reasons for the additional image-based inspection.

The first step was to identify the optimal assistance system. Assembly assistance systems available on the market could not meet Endress+Hauser's requirements. One reason was that these systems are often cloud-based, which was not an option for Endress+Hauser due to data protection concerns and availability considerations.

An alternative would have been extensive development work to integrate such systems into the MES. However, additional third-party hardware would have introduced further complications. The simplest and most efficient solution turned out to be the integration of standard machine vision software.

"Machine vision enables fast, contactless testing in real time. At the same time, our production control system already has numerous interfaces that we could use to integrate image processing software, allowing it to run on hardware managed by us," Krause explains.

The software selected was MVTec MERLIC. MERLIC is a no-code software that enables users to create complete machine vision applications without programming knowledge. In addition, the software includes numerous powerful tools that can solve almost all machine vision tasks.

Information for the assembly process

The assembly workstation where the machine-vision application is used consists of a height-adjustable table made of Bosch profiles with storage compartments. Here, a camera is mounted on the upper rail. This camera accompanies the employee during the assembly of the components.

These components are typically flowmeters nearing completion, and the process steps are as follows: first, the code that accompanies the components is read. A screen at the workstation then displays messages and images showing how the component must be assembled and what needs to be considered. This ensures that critical steps are carried out correctly and consistently.

MERLIC is also used as part of quality assurance. For example, the software detects whether seals have been installed. This step is crucial for ensuring product quality in the manufacturing process. The correct execution of these critical process steps is reported to the MES, and the device is released for the next process step.

Simple integration into the assembly system

"We chose MERLIC for a number of reasons. The most important is that, as a standard software, it serves as a tool we can use for various tasks. As a result, we do not have to train our employees on different types of software. Finally, process integration is straightforward. MERLIC's numerous open interfaces - such as GenICam - give users great flexibility regarding compatible hardware and make software integration simple," explains Julius Krause.

At the workstation, MERLIC is embedded in a streamlined machine vision setup. This includes an industrial PC, a laser distance sensor for distance measurement, and a camera with a liquid lens for variable focus. Machine vision software forms the core of a machine vision application, handling the processing and evaluation of captured images.

Barcode scanners are used to scan the codes on the components. The information behind these codes includes how the present component with a batch size of 1 must be assembled step by step and what needs to be considered. To ensure an efficient workflow, the camera and MERLIC are running in the background, automatically recognising and confirming correct execution so that the user interface (UI) advances to the next step.

Two AI-based technologies included in MERLIC are used for recognition and confirmation: semantic segmentation and object detection. Semantic segmentation enables pixel-precise localisation of error classes, while object detection locates object classes and identifies them using bounding boxes.

"The application at Endress+Hauser clearly demonstrates how machine vision can easily help customers digitize their production while increasing quality. As a no-code software, MERLIC is ideally suited to work hand in hand with employees - simply because it is easy to use and can therefore be flexibly applied to various machine vision tasks," explains Ulf Schulmeyer, Product Manager for MERLIC at MVTec.

An application that will be expanded further

Since quality control with MERLIC met the requirements, Endress+Hauser is currently moving to the next stage of expansion. In this phase, MERLIC will take on additional tasks. The goal is for MERLIC to independently read the codes on components using machine vision technologies such as barcode reading and OCR (optical character recognition), and to retrieve the digitally stored information behind them. At the start of their work, the production employee will only need to hold the workpiece under the camera so that the code can first be read and the workpiece can later be inspected during the process.

At Endress+Hauser, great importance is placed on always using the latest techniques and technologies, including in assembly. The company has developed a structured approach to developing, testing, validating, and ultimately rolling out new manufacturing processes worldwide. Development, testing, and validation take place at the company's headquarters in Reinach, Switzerland - including the machine-vision-assisted assembly process.

"We have been using MERLIC since March 2025. The software proved to be stable and reliable. That is why we began rolling out this process to our other global sites in fall 2025," says Julius Krause.

For seamless integration of the software into the assembly process, Endress+Hauser worked with MVTec's customer support team. "We had some feature requests to optimise the system. For example, the MVTec team optimised the export of AI models to the target hardware and developed a communicator plugin that sends images directly to our central servers," Julius Krause concludes.