Fraunhofer Institute for Industrial Mathematics ITWMWhen does Quantum Computing offer a real advantage for industrial applications, and when are classical methods the better choice? With »QUALLE« we at Fraunhofer ITWM are developing an agent-based benchmarking platform that systematically investigates this question. Companies provide their existing workflows and QUALLE compares suitable classical, quantum-inspired and hybrid quantum-classical methods to determine which approach is best suited for the task at hand.
Keyvisual »QUALLE«: QUantum Advantage? Let's Look Empirically
»QUALLE« compares classical, quantum-inspired and quantum methods for specific industrial applications and demonstrates which approach performs best under realistic conditions.
© magnific
Quantum Computing for Industrial Applications: From Industrial Workflow to Quantum Fit
Project »QUALLE«: QUantum Advantage? Let's Look Empirically
The central idea is already embedded in the title »QUantum Advantage? Let’s Look Empirically«: Rather than assuming a quantum advantage exists, we want to investigate it empirically and make it measurable.
Your Own Workflow Becomes the Standard
The starting point is always a specific real-world challenge, such as a quality inspection, an optimization, or a simulation. The existing industrial workflow serves as a benchmark. All alternative processes must be evaluated against it.
An AI agent transfers the workflow to the high-performance computers at Fraunhofer ITWM and suggests suitable methods from a growing library. After approval –by a human– it performs the comparisons on the same data and under identical conditions. In doing so, it uses the appropriate hardware in each case, ranging from CPU and GPU systems to quantum simulators and Quantum Computers.
Ultimately, »QUALLE« does more than just show which method performs best. The platform also highlights the quality of results, processing time and costs associated with each method, as well as why a particular approach is suitable for a given task.
This creates a robust path from today's industrial workflows to Quantum Computing. Companies can identify where quantum methods already offer tangible added value and where classical solutions continue to have the advantage.
Significant Is the Application as a Whole
Our research shows that whether a quantum method is worthwhile cannot be determined based on a single algorithm. The overall application is what matters. Data preparation, encoding, training, and evaluation influence the result just as much as the actual quantum component.
»QUALLE« therefore takes the entire workflow into account. This makes it possible to compare different technological approaches under realistic conditions.
How Companies Benefit from »QUALLE«
The Right Approach for Each Workflow: We don’t provide companies with a blanket statement about whether Quantum Computing is »better.« Instead, we compare classical, quantum-inspired and quantum approaches based on the specific task, the available data and the existing infrastructure. We provide guidance on how to optimize these processes.
Low barrier to entry: Users do not need in-depth knowledge of either Quantum Computing or high-performance computing. The agent transfers the workflow, suggests appropriate methods, schedules the calculations and presents the results in an easy-to-understand format.
Quantum Advantage, Not Just Hype: Quantum methods compete against powerful classical methods under comparable conditions. Only this direct comparison reveals whether –and under what conditions– an actual advantage arises.
The Right Hardware for Every Task: QUALLE can run the same analysis on different target systems. This reveals how the quality of results, runtime and costs differ across CPUs, GPUs, simulators and quantum hardware. Quantum Computers from different manufacturers can be connected via a common interface.
Traceable results: The data set, parameters, software environment and hardware used are documented for each run. At the same time, you can control how much Computing time is available and whether external AI models are allowed to access project data.
Developing New Quantum Methods With AI Agents
Agents can not only compare existing methods but also design the quantum component of a method themselves. In the case of a hybrid classifier, for example, an agent can design a quantum circuit, test it and refine it based on the results. In doing so, the agent can also take other designs into account or adopt a circuit from a scientific publication and investigate whether the published results can be reproduced.
Each new design must then undergo the same tests as existing methods. To do this »QUALLE« compares the quantum circuit, among other things, with variants that do not use entanglement, with random parameters instead of trained ones, or with a classical layer of comparable size. This makes it possible to verify whether the quantum component actually contributes to the result.
Under clearly defined security conditions, agents can also create their own program code and deploy their designs independently. The code runs in isolation, without network access, in verified software environments and within a predetermined computing budget. Each design is documented in a traceable manner, including its origin and the corresponding assignment.
Multiple agents can also work on the same task collaboratively. In doing so, each agent pursues its own solution path. All agents work with the same task description, the same evaluation criteria and a fixed computational budget. »QUALLE« then compares the results and can combine particularly successful approaches.
Case Study: Automated Inspection of Welds
Quality control of aluminum welds produced using the tungsten inert gas (TIG) process illustrates how this works in practice. Camera images of the welds are classified into six categories: a defect-free weld and five different types of defects.
A traditional method first describes each image using 2,268 features and then classifies it into one of six classes using a support vector machine. However, to ensure that quality inspection can be performed as quickly as possible directly on the production line, a significantly smaller subset of these features should suffice.
The search for the best features is what is known as a combinatorial optimization problem: from a very large number of possible combinations, the goal is to find the one that is particularly meaningful for image recognition. »QUALLE« translates this task into a QUBO model (Quadratic Unconstrained Binary Optimization), a mathematical formulation that can be solved using various classical and quantum methods.
Case Study: Weld Inspection: »QUALLE« is investigating various methods for reliably detecting defects in aluminum welds produced using the tungsten inert gas (TIG) process.
Next, various solution approaches are pitted against one another. These include the classic tabu search, quantum-inspired methods such as simulated quantum annealing and simulated bifurcation on graphics processing units, as well as QAOA (Quantum Approximate Optimization Algorithm), an algorithm for Quantum Computers. All methods use the same input data. The selected features are then tested in the classifier. This allows for a direct comparison of which approach achieves the best results under the given conditions.
The results also demonstrate why such a technology-neutral comparison is important: Selecting less suitable features can speed up the traditional testing process. However, in this example, a neural network running on a GPU achieves better results in detecting errors.
This is where »QUALLE« really shines: The platform does not assume a quantum advantage, but rather makes it possible to determine which method is best suited for a specific task and the available hardware. This provides a transparent and reproducible basis for deciding between classical, quantum-inspired, and quantum methods.
Comparable Results on Different Hardware
To ensure that the results remain meaningful, »QUALLE« documents, among other things, the pipeline version, the settings of the individual modules, the data status, the software environment and the computing resources used for each run. Only calculations performed under comparable conditions are evaluated together. Repeated runs with different random values are statistically incorporated into the evaluation.
The platform has a modular design. A workflow consists of interchangeable building blocks and “QUALLE” checks how they interact before the workflow starts. This makes it relatively easy to integrate new methods, quantum backends, simulators, or additional hardware.
Currently, more than 80 modules are available. Among other things, they cover classification and regression using classical machine learning, neural networks, hybrid quantum circuits, quantum kernel methods, and tensor networks, as well as data preparation, optimization, and quantum error correction.
For its calculations, the platform uses the appropriate resources in each case, ranging from individual processor cores to graphics processing units and multiple computing nodes, all the way to quantum simulators and Quantum Computers from various manufacturers. A browser-based interface displays pipelines, calculations and comparisons. Roles and permissions determine who within a project is allowed to view data, make changes or start calculations.
Objectively Evaluate Quantum Methods and Accelerate Research
»QUALLE« also provides a common basis for comparison in research. Researchers can integrate new methods as modules or incorporate new quantum circuits into existing methods. These can then be compared with established alternatives under identical conditions. Results from scientific publications can also be reproduced in this way and directly compared with the researchers’ own approaches.
In this way, »QUALLE« helps distinguish verifiable progress from unrealistic expectations. At the same time, it makes it possible to determine what steps are necessary to turn a promising quantum algorithm into a solution for industrial applications.