BQP (BosonQ Psi) and materialsIN, two deep-tech ventures based in Upstate New York, have partnered on a novel Quantum Machine Learning (QML) solution that addresses 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, www.materialsin.com, a materials informatics venture with extensive expertise in materials science and advanced informatics techniques, provides innovative data-driven solutions 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 BosonQ Psi to translate into quantum terms.
BosonQ Psi, www.bosonqpsi.com, 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.
BosonQ Psi 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.
Problem Statement and Motivation
BQP and materialsIN were engaged to solve a real-time problem in the concrete space-existing concrete infrastructure, which is continually exposed to extreme environments, requiring continual maintenance. The underlying cause of critical structural failure can be avoided through early assessment of the presence of defects and cracks. Since surface anomalies defects can be visually examined, a variety of computer vision methodologies were applied to assess the structure's integrity. The latest iteration of this approach uses a convolutional neural network (CNN) to achieve state-of-the-art performance. In this exercise, we compared the performance of the state-of-the-art conventional neural network methods to next-generation quantum neural network methods.
Materials Informatics, a data-centric approach to materials engineering, is rapidly evolving to advance materials discovery, characterization, design, synthesis, and screening of new or alternative materials [1,2]. The advantages of leveraging vast datasets and advanced computational techniques can accelerate these processes. However, the complexity and heterogeneity of materials data, coupled with the limitations of traditional machine learning models, pose significant challenges. BQP’s approach incorporates a hybrid quantum-classical machine learning model to accurately identify cracks, even in challenging scenarios. Accurate crack detection in infrastructure materials, like asphalt and concrete, is crucial for safety. However, traditional methods struggle with complex crack patterns and varying environmental conditions (i.e., lighting, shadows, and textures), demanding advanced AI solutions. Quantum machine learning (QML) can advance image classification to identify rare defects and improve overall accuracy.
Quantum computing with AI provides transformative advancements in material science. The Department of Energy has used quantum annealing to simulate magnetic materials with a level of detail that mirrors real-world experiments. By leveraging AI to predict material properties and quantum computing to perform dense functional theory (DFT) calculations faster than traditional methods, Microsoft’s AI has screened 32 million efficient and environmentally friendly battery materials. Moreover, research is conducted to develop new carbon-based quantum materials for developing high-speed electronic devices at the molecular level and creating quantum bits. Quantum computers use several attributes such as superposition, entanglement, interference, and tunneling that require careful manipulation and control to be utilized effectively in quantum computing to achieve unimaginable speed. AI problems, such as image classification, use these attributes to exploit quantum parallelism to process vast datasets simultaneously. Research has shown that QML models can effectively handle imbalanced classification problems, such as medical imaging where minority classes are critical.
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 aligned 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.
Figure 1: Representation of the Hybrid Quantum Transfer Learning Process
Data description
The dataset used for this research contained 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.
Chart 1: Comparison of trainable parameters
Chart 2: Comparison of accuracy
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, delivering a remarkable 99.8% accuracy compared to the classical model's 98.5%. The F1-score, a harmonic mean of precision and recall, further emphasizes the quantum model's superiority with a value of 0.9921 compared to the classical model's 0.9590. These results indicate that the quantum model consistently performs better across different class imbalance levels.
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.
Conclusion
This convergence of QML and Materials Informatics offers transformative results. This use case on anomaly detection demonstrates QML's effectiveness in handling imbalanced datasets prevalent in a wide range of industries, such as aerospace, defense, and automotive. The BQP-materialsIN hybrid model surpasses traditional methods by addressing data challenges and achieving computational efficiency, and its versatility positions it as a valuable tool for various applications. The collaboration between materialsIN and BQP showcases the power of combining classical and quantum computing, and, by building on this foundation, the partners can unlock QML's full potential to address a broad range of problems faced by industries.
While this use case offered promising results, further investigations are necessary to quantify the exact parameter reduction, assess the model's performance on edge devices, and explore its applicability to a wider range of problems. By deepening the partnership, BQP and materialsIN can pave the way for a future where quantum machine learning becomes an indispensable tool for addressing global challenges.
Future Outlook
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.
References:
Comments