BQP and materialsIN, two deep-tech ventures ​​based in​​ Upstate New York, have partnered on a novel Quantum Machine Learning (QML) solution that​​​​​​​​​​ address​​​es​​​ the challenges associated with Material Informatics using classical Machine Learning. ​​

​​​The partners were recently engaged to apply their approach and solution to​​​​​​ a live use case: surface crack detection of concrete materials. ​​​The result was a solution that yielded higher accuracy, at reduced time and cost.​​​​​

​​​materialsIN gathered/curated a comprehensive dataset (​​​comprised of ​​​images), which it forwarded to BQP to utilize in its simulation platform, BQPhy®’s ML library, with the mission to improve the training of the advanced quantum machine learning model for its automatic crack detection use case.  ​​

MatetrialsIN, a materials informatics venture with extensive expertise in materials science and advanced informatics techniques, provides innovative data-driven solution​s​ in ​various​ domains. The methodology allows for sophisticated modeling and simulation capabilities, crucial for the predictive analytics of materials. In this use case, materialsIN used its data-driven, machine learning approach to define the problem and curate the data for BQP to translate into quantum terms.  

BQP, a quantum-powered simulation venture, then applied its engineering simulation platform. Its quantum algorithm-based solver creates simulations that are 10x cost-effective, faster, and more accurate.  

BQP and materialsIN leveraged their respective domain expertise to arrive at an expedient, cost-effective, and highly accurate solution to help manufacturers with their materials utilization needs. ​The partnership enables manufacturers to produce higher quality materials more efficiently and cost-effectively, resulting in a reduction of $1-2Bn and slashing 2-3 years from product cycles.​

What Is Quantum Machine Learning and Why Does It Matter for Materials Science?

Quantum Machine Learning (QML) combines quantum computing's ability to process multiple states simultaneously with classical machine learning frameworks. Instead of replacing classical models, QML enhances them using quantum layers to handle high-dimensional, complex pattern recognition that classical networks approximate at significantly higher computational cost.

In materials science, where datasets are heterogeneous, imbalanced, and often sparse, this distinction matters. QML models generalize better from limited data, converge faster, and maintain accuracy under conditions that cause classical models to degrade.

Why It Matters for Materials Science:

• Handles data heterogeneity: Materials datasets vary widely in texture, resolution, and surface conditions QML models process this variance without requiring extensive augmentation or preprocessing.
• Performs on imbalanced datasets: Defects and cracks are rare events by nature. QML maintains high precision and recall even when positive cases represent a small fraction of total samples.
• Reduces parameter overhead: BQP's hybrid model achieved superior accuracy with 2,137 parameters versus over 14 million in classical models directly lowering compute cost and training time.
• Generalizes from limited data: In materials research, failure data is often scarce. QML models reach near-perfect accuracy on datasets as small as 1,000 images where classical models show instability.
• Accelerates defect detection cycles: Faster convergence means shorter iteration loops between data collection, model training, and deployment compressing materials characterization timelines.
• Bridges current hardware constraints: Hybrid architectures align with NISQ-era limitations, making QML deployable today without waiting for fault-tolerant quantum hardware.

Quantum Machine Learning: Problem Statement and Motivation

Existing concrete infrastructure is continually exposed to extreme environments, making early defect detection critical to preventing structural failure. Traditional computer vision methods  including state-of-the-art CNNs struggle with complex crack patterns, varying lighting conditions, shadows, and surface textures. BQP and materialsIN were engaged to test whether quantum neural network methods could outperform these classical approaches on a real dataset.

The Core Problem:

• Surface cracks in concrete are visually detectable but computationally difficult to classify reliably under varying environmental conditions.
• Classical CNNs achieve strong baseline accuracy but struggle with imbalanced datasets where crack instances are rare relative to non-crack samples.
• Materials datasets are heterogeneous high variance in surface finish, resolution, and lighting makes consistent model performance hard to maintain.
• Traditional ML models require large, balanced datasets to generalize well, which is rarely achievable in real-world materials inspection scenarios.

Why Quantum Machine Learning?

• Materials Informatics is evolving rapidly, but classical ML limitations slow down materials discovery, characterization, and defect screening at scale.
• QML exploits superposition, entanglement, and quantum parallelism to process high-dimensional image data more efficiently than classical approaches.
• QML models have demonstrated effectiveness on imbalanced classification problems such as medical imaging and defect detection where minority classes are critical.

The Department of Energy has used quantum annealing to simulate magnetic materials; Microsoft's AI screened 32 million battery material candidates using quantum-accelerated DFT calculations establishing real precedent for quantum methods in materials science.

Architecture  

In this use case, ​the partners ​developed a hybrid quantum neural network that combines the strengths of classical and quantum computing. Additionally, they used​​ a pre-trained classical model as feature extractors. These extracted features ​were ​then transferred to a quantum​​ layer for further processing, and​​ align​ed​ with current NISQ hardware limitations. In turn, they offered immense potential for their applications in surface defect and cracks detection. The model achieved improved accuracy and efficiency, showcasing the potential of QML for complex problems, especially in this image-processing use case. Importantly, it addresses scalability issues associated with more complex problems.

Data Description

The dataset used for this research contain​ed ​40,000 images of concrete surfaces, divided into 'negative' (without cracks) and 'positive' (with cracks) classes. Each image is a 227x227 pixel RGB image. The dataset ​was​ comprised of high-resolution images, exhibiting significant variance in surface finish and lighting conditions. To maintain data consistency, no augmentations were applied.

Result description  

BQP and materialsIN applied the hybrid quantum convolutional neural network (HQCNN) ​to analyze​ the dataset of high-resolution RGB high-accuracy images provided by materialsIN, capturing the detailed textures and contours essential for differentiating cracks from other similar patterns in construction materials for identifying these cracks. The comparison of our HQCNN with VGG16 and Low-Rank Approximation (LoRA) classical network architecture models showcased the advantages of the hybrid model developed by BQP.  

LoRA is a technique ​that is ​used to efficiently fine-tune neural networks by introducing small and low-rank matrices to the neural network, rather than modifying the entire model’s weight. This approach significantly reduced​​​​ the number of trainable parameters, making the fine-tuning process more efficient.

Classical Model Performance

The classical model used 14714688 trainable parameters while attaining an accuracy of 93.44%. It has struggled to accurately predict positive cases.  

​​Hybrid Quantum Neural Network Model Performance

Our hybrid model outperformed the classical approach across all evaluation metrics with only 2137 trainable parameters, while achieving 98% accuracy.

Impact of Imbalanced Datasets on Model Performance

Dataset (10:90 Crack:Non-Crack)

Our hybrid model excelled in handling the highly imbalanced dataset, achieving a remarkable 99.8% accuracy compared to the classical model's 98.5%. The F1-score, which balances precision and recall, further highlights the quantum model's advantage with a value of 0.9921 versus the classical model's 0.9590. These results demonstrate that the quantum model consistently outperforms across varying levels of class imbalance. Such performance gains underscore the potential of Quantum Evolutionary Algorithms in Aerospace, where precise classification and optimization under complex, imbalanced conditions are critical for mission-critical systems and design innovations.

Dataset (70:30 Crack:Non-Crack)

While the class imbalance was less severe in this dataset, our hybrid model consistently outperformed the classical approach, achieving 99.55% accuracy versus 97.87%. The F1-score also favors the quantum model with a value of 0.9970 compared to the classical model's 0.9846. This reinforces the hybrid model's adaptability to varying imbalance levels.

Impact of Small Dataset on Hybrid Model

Our hybrid model demonstrated exceptional performance even with a relatively small dataset of 1000 images, consistently surpassing the classical model in accuracy. The hybrid model exhibited rapid convergence, reaching near-perfect accuracy (99% after 10 epochs, 100% after 20 epochs). In contrast, the classical model peaked at 98.73% accuracy (after 20 epochs) and displayed instability at earlier stages (95.67% after 10 epochs). Classical Model has obtained an F1 score of 0.9876, which is very good but not perfect; however, the quantum model achieved a perfect score of 1.000, indicating that the model correctly identified all positive instances without any false positives or false negatives.

Why Classical Machine Learning Falls Short for Crack Detection

Classical CNNs set the benchmark for image-based defect detection but in real-world materials inspection, they hit consistent limits. The problems aren't edge cases; they're structural to how classical models process information.

Where Classical Models Break Down:

• Parameter bloat: VGG16 requires over 14 million trainable parameters to achieve 93.44% accuracy making training computationally expensive and slow to iterate.
• Imbalanced dataset sensitivity: When crack instances are rare (10:90 split), classical models struggle to accurately predict positive cases, producing high false-negative rates that miss real defects.
• Limited generalization on small data: Classical models peak at 98.73% accuracy on 1,000-image datasets and show instability at earlier training stages unreliable for inspection scenarios with limited labeled data.
• Environmental noise sensitivity: Varying lighting, shadows, and surface textures degrade classical model performance without extensive augmentation and preprocessing pipelines.
• Slow convergence: Classical models require more epochs to stabilize, increasing training time and infrastructure cost across iterative development cycles.

Where the Quantum Convolutional Layer Changes the Equation:

• Exponentially fewer parameters: BQP's HQCNN achieves 98% accuracy with just 2,137 trainable parameters a 6,800x reduction over the classical model with higher accuracy.
• Superior imbalance handling: The quantum layer maintains an F1 score of 0.9921 on a 10:90 split, versus 0.9590 for the classical model  directly reducing missed defect detections.
• Rapid convergence on small datasets: The hybrid model reaches 99% accuracy after 10 epochs and 100% after 20 compared to 95.67% and 98.73% respectively for the classical approach.
• Physics-aligned feature processing: The quantum layer processes extracted classical features through quantum state transformations, capturing correlations in high-dimensional data that classical layers approximate less precisely.
• NISQ-compatible deployment: The architecture is designed within current hardware constraints making it deployable today without requiring fault-tolerant quantum systems.

What These Results Mean for Industry Decision-Makers

The numbers from this study aren't just benchmarks they translate directly into inspection reliability, operational cost, and risk exposure for engineering teams deploying defect detection at scale.

Higher Accuracy Means Fewer Missed Defects

The hybrid model's 98–99.8% accuracy across dataset conditions isn't an incremental gain it closes the gap between laboratory-grade inspection and real-world deployment reliability. In structural and manufacturing contexts, a missed crack isn't a model error; it's a safety or quality failure with downstream cost consequences.

Fewer Parameters Mean Lower Compute Cost

Running a 2,137-parameter model versus a 14-million-parameter classical network at production scale is not a marginal efficiency difference. For organizations running continuous inspection workflows across large asset bases semiconductor fabs, aerospace MRO facilities, civil infrastructure the infrastructure cost reduction is direct and compounding.

Small Dataset Performance Removes a Key Adoption Barrier

Most industrial inspection environments don't have millions of labeled defect images. The hybrid model reaching 100% accuracy at 20 epochs on 1,000 images means organizations can deploy effective QML-based inspection without first investing years in dataset construction a barrier that has historically slowed AI adoption in manufacturing and materials science.

Imbalanced Data Handling Aligns with How Defects Actually Occur

Defects are rare by definition. A model that degrades on imbalanced datasets is a model that fails precisely when defects matter most. The hybrid model's consistent F1 advantage across 10:90 and 70:30 splits means performance holds in real inspection conditions, not just controlled test environments.

Cross-Industry Applicability Reduces Implementation Risk

The same architecture validated on concrete crack detection applies directly to semiconductor wafer inspection, aerospace structural monitoring, and automotive surface quality control. Organizations in these sectors can adopt a proven model architecture rather than building custom solutions from scratch reducing both development time and deployment risk.

Conclusion

This use case demonstrates that quantum machine learning isn't a future capability it's a present one. The BQP-materialsIN hybrid model consistently outperformed classical CNNs across accuracy, parameter efficiency, imbalanced data handling, and small dataset performance. For industries where defect detection directly affects safety, yield, and cost, these gains are operationally significant.

The results also point to something broader. The same architectural approach that outperformed VGG16 on concrete crack detection applies to semiconductor inspection, aerospace structural monitoring, automotive quality control, and pharmaceutical imaging. Organizations don't need to build QML capabilities from scratch a validated, production-aligned architecture already exists.

BQP's BQPhy® platform is the engine behind this hybrid approach. By combining quantum-inspired algorithms with classical feature extraction, BQPhy® delivers simulation and detection capabilities that classical tools cannot match running on existing HPC and GPU infrastructure, without quantum hardware dependencies. Whether the challenge is materials defect detection, multi-physics simulation, or large-scale engineering optimization, BQP provides the computational layer that closes the gap between where classical methods fall short and where outcomes need to be.

Further investigations are underway to quantify exact parameter reduction at greater scale, assess model performance on edge devices, and extend the architecture to broader problem classes including Topology Optimization of Airfoil designs for aerodynamic performance and fuel efficiency.

Quantum Machine Learning Future Trends

The model's efficiency and accuracy make it suitable for a wide range of industrial applications:

Manufacturing

• Optimize Product Scheduling: reduce lead times and minimize idle resources.
• Predictive maintenance: prevents unplanned downtime through equipment failure prediction.
• Supply chain optimization improves inventory management and demand forecasting.

Semiconductor

• Improve yield: early defect detection reduces waste and costs.
• Enhance quality control: precise defect classification improves chip reliability.
• Predictive maintenance: prevent defects through early issue detection.
• Accelerate design optimizes chip layout and design for faster time-to-market.

Aerospace and Defense

• Structural health monitoring improves aircraft safety and reduces maintenance costs.
• Target recognition: enhance weapon system effectiveness.
• Autonomous systems: enables drones and vehicles for various missions.

Automotive

• Autonomous driving: enable vehicles to perceive and navigate environments independently.
• Predictive maintenance: optimize vehicle lifespan through early failure detection.
• Advanced driver assistance: enhance road safety with features like lane keeping assist.

Energy

• Demand forecasting: optimize power generation and distribution.
• Renewable integration: improve grid efficiency and reliability.
• Energy efficiency: identify opportunities to reduce energy consumption.

Pharmaceutical

• Drug discovery: accelerate drug discovery by analyzing vast amounts of data to identify potential drug candidates.
• Personalized medicine: develop personalized treatment plans, improving patient outcomes.

FAQs

1. What is Quantum Machine Learning and how is it different from classical ML? 

Quantum Machine Learning combines classical machine learning frameworks with quantum computing layers that process information using superposition, entanglement, and quantum parallelism. Unlike classical models that evaluate features sequentially, QML processes high-dimensional data more efficiently achieving higher accuracy with significantly fewer trainable parameters, as demonstrated in BQP's hybrid model.

2. Why does the hybrid quantum-classical model outperform classical CNNs for crack detection? 

Classical CNNs require millions of parameters to handle complex, variable surface conditions and still struggle on imbalanced datasets. BQP's Hybrid Quantum Convolutional Neural Network (HQCNN) uses a quantum layer to process extracted classical features more efficiently achieving 98% accuracy with just 2,137 parameters versus over 14 million in the classical model, while maintaining superior F1 scores across imbalanced and small datasets.

3. Does deploying QML require quantum hardware? 

No. BQP's hybrid architecture runs on existing HPC and GPU infrastructure. The quantum convolutional layer is designed within NISQ-era constraints, meaning organizations can deploy quantum-enhanced defect detection today without specialized quantum hardware, cryogenic cooling, or cloud QPU access.

4. Which industries can benefit from BQP's QML-based defect detection approach? 

The HQCNN architecture validated on concrete crack detection applies directly across semiconductor wafer inspection, aerospace structural health monitoring, automotive surface quality control, energy infrastructure assessment, and pharmaceutical imaging. Any inspection workflow dealing with rare defects, imbalanced datasets, or limited labeled data is a strong candidate.

5. How does BQP's platform handle scenarios where defect data is scarce? 

BQP's hybrid model reached 99% accuracy after just 10 training epochs and 100% after 20 epochs on a dataset of 1,000 images outperforming classical models that showed instability at the same stage. This makes BQPhy® deployable in real industrial environments where large labeled defect datasets are not available, removing one of the primary barriers to AI adoption in materials inspection.