Vision For The Future of Manufacturing – Inspection Data To Create Robot Rework Programs

Imagine a world where manufacturing becomes smarter, more efficient, and remarkably innovative. It’s no longer just a vision; it’s a reality with the unveiling of the groundbreaking ‘SeConRob’ project (Self-configuring Multi-Step Robotic Workflows). This exciting initiative, supported by Horizon Europe, is set to transform the manufacturing landscape with a budget of almost 3 million EURO. The key players behind the innovation include PROFACTOR GmbH (Austria), Safe Metal (France), Otto Fuchs (Germany), ACS (Germany), ECL (France), Fraunhofer IZFP (Germany) and Marposs (Italy).

In many production processes there are situations in which the subsequent process step depends significantly on the results of the previous process step. A typical example is the combination of quality assurance and rework. As part of quality assurance, potential defects are found that need to be remedied through a rework process. The specific execution of the rework depends entirely on the results of quality assurance; the position, type and severity of the defects found influence the entire process. Automating such processes is currently relatively difficult because there are no suitable options for automatically configuring more complicated processes.

In the first step, for example, a fluorescence test (MPI) is carried out, which indicates inhomogeneities (cracks) on or below the surface. Depending on the position and size of the defect, material is selectively removed until the defect is completely removed. The area must then be filled again using a welding process. In order to produce a sufficiently high surface quality, local grinding and polishing is carried out at the end. This process is now carried out completely automatically at the company partner Safe Metal, for example, because automation is currently not possible.

Manufacturing processes often involve scenarios where the success of a subsequent step hinges on the results of the preceding one. A classic example is the fusion of quality assurance and rework. During quality assurance, potential defects surface, necessitating a rework process for correction. However, the intricacies of this rework rely entirely on the quality assurance results, including the defect’s location, type, and severity, all of which significantly impact the overall process. Automating such processes has been a formidable challenge due to the complexity of configuring intricate processes automatically.

As part of the project, it is assumed that the manual processes used so far are fundamentally suitable for the tasks. The focus is therefore on the automation of the processes and especially on the automatic configuration of a multi-stage chain of process steps. The following sub-technologies are necessary for this:

Parametric, physical process models that allow automatic planning of a robot process and can be configured based on information from previous process steps.

Methods for data analysis (sometimes using AI) that extract additional information from data from measurement and quality assurance systems that is necessary for configuring subsequent processes. Typically, such information goes well beyond the simple good/bad decision.

Longer-term feedback loops (e.g. based on reinforcement learning) with which the configuration of the processes can be gradually improved so that the quality of the components improves or fewer iterations are necessary to achieve a defect-free component.

As part of the project, two application cases from the processing and testing of metallic castings and forgings are dealt with. In this context, two robot cells will be set up to demonstrate the automatic configuration of an entire process chain in a realistic environment.

‘SeConRob’ embarked on its journey on October 1, 2023, with a kick-off meeting at PROFACTOR. The next steps are already planned, which include documenting use cases, defining processes, and identifying critical process parameters. The planning of the two robotic cells and all necessary hardware components is also on the horizon, along with experiments to determine process parameters for processes that were previously untouched by automation.

The companies involved, Safe Metal and Otto Fuchs, operate in the realm of safety-critical components and aviation, where stringent quality standards and rework processes are the norm. The current processes are neither automatable nor scalable to meet the projected aviation production volumes. This bottleneck, attributed to limited automation capabilities and a shortage of highly skilled personnel, is what ‘SeConRob’ aims to eliminate. “We’re offering a forward-looking solution that paves the way for these companies to thrive. Moreover, Marposs (Inline Process Monitoring) and ACS (Robotic Ultrasonic Testing Systems) are set to expand their technological applications into exciting new domains through this project.”

Inspection Data To Create Robot Program

The Fraunhofer IZFP will perform AI-driven data analysis to extract information from inspection data to create a robot program and generate process parameters for the downstream finishing process. Based on physical process models, a long-term feedback loop will be established to optimize the process and consider features not included in the original model.

In this context, Fraunhofer IZFP will develop AI-based software for defect detection and segmentation. For both use cases, this includes the development of suitable models for the data generated by the trials at the process level and at the intermediate demonstration level. At the same time, machine learning methods will be used to classify different types of defects as required by subsequent process steps.

Subsequently, all segmented regions will be characterized by features that are needed to set up the next downstream process. Candidates for such features are the size of the defects or the position under different coordinate systems of the inspection sensors. The feature extraction is linked to the data flow along the process steps. The next step is to create structured datasets suitable for modelling, training, validating, and benchmarking the AI methods.

Datasets of different size and quality will need to be created to allow for continuous improvement along the different levels of automation, starting with process level trials and ending with multi-level trials. Data augmentation methods will be used to minimize the size of the datasets and to generalize the models for different shapes and varying conditions of the processes.

‘SeConRob’ isn’t just a project; it’s a vision for the future of manufacturing.