QSVM-RQNN: Low-Qubit Recurrent Quantum Similarity Learning for Condition Monitoring and Fault Classification

Industrial condition monitoring requires high-dimensional data processing, which is often difficult for current Noisy Intermediate-Scale Quantum (NISQ) devices due to limited qubit counts and hardware noise. The authors address this by proposing QSVM-RQNN, a hybrid framework that combines Quantum Support Vector Machine (QSVM) similarity learning with Recurrent Quantum Neural Networks (RQNNs).

The approach follows a four-step pipeline:

1. Dimensionality Reduction: Principal Component Analysis (PCA) compresses high-dimensional sensor data.
2. Data Balancing: The Quantum Synthetic Minority Oversampling Technique (QSMOTE) addresses class imbalance.
3. Similarity Learning: The model computes similarity between input features and class centroids.
4. Recurrent Processing: A compact three-qubit quantum circuit with shared parameters processes the data sequentially to classify faults.

Two variants were developed: V1 performs joint encoding of inputs and centroids before recurrent processing, while V2 computes similarity features at each timestep before feeding them into the recurrent network.

The QSVM-RQNN framework demonstrates that combining similarity-based learning with recurrent quantum architectures allows for expressive sequence modeling without requiring large-scale quantum resources. Experimental results across multiple fault diagnosis datasets show that the model achieves competitive or state-of-the-art performance compared to standalone QSVM, QNN, QCNN, and other hybrid models. The architecture is particularly effective at improving recall and classification accuracy on highly imbalanced datasets, which are common in industrial predictive maintenance.

This research provides a scalable path for deploying quantum machine learning in industrial settings. By utilizing parameter sharing and a fixed three-qubit structure, the framework mitigates the need for deep, noise-sensitive circuits. This makes it a practical candidate for real-time condition monitoring on near-term quantum hardware, where qubit availability remains the primary bottleneck for complex AI tasks.