Granulation Fault Diagnosis Using VAE and Contribution Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional monitoring methods struggle to handle the nonlinearity and non-Gaussianity of data in continuous pharmaceutical processes, limiting their ability to accurately extract features and diagnose faults in drug particle production.
Innovation Solution
A monitoring and fault diagnosis method using a variational autoencoder combined with a contribution graph, which adaptively extracts complex nonlinear features and maps high-dimensional data to low-dimensional latent space, enabling precise fault detection and diagnosis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional multivariate statistical process monitoring methods (PCA, PLS) are used, then the monitoring system is simple and easy to implement, but they cannot effectively handle the nonlinearity and non-Gaussianity of data in continuous pharmaceutical processes
Solution Approach 1:
The patent transforms the monitoring approach by changing from linear statistical parameters (PCA, PLS) to nonlinear deep learning parameters (autoencoder reconstruction error, KL divergence). This parameter transformation enables the system to capture non-Gaussian characteristics and nonlinear relationships in granulation process data, significantly improving monitoring accuracy while maintaining computational feasibility.
Solution Approach 2:
The patent replaces conventional statistical mechanics (multivariate statistical methods) with intelligent mechanics (deep learning-based variational autoencoder). This substitution allows the system to automatically learn complex nonlinear patterns and non-Gaussian distributions in the data, resolving the limitation of traditional methods while maintaining system implementability.
2Reliability
If kernel methods (KPCA, KPLS) are applied to handle nonlinear processes, then the monitoring performance improves, but they are only applicable to small and medium samples due to kernel technology limitations
Solution Approach 1:
The patent extracts the essential feature of handling nonlinear and non-Gaussian data by using autoencoder-based feature representation. By focusing on reconstruction error and latent space distribution characteristics, the method extracts discriminative features that work effectively across different sample sizes, eliminating the sample size constraint of kernel methods.
Solution Approach 2:
The variational autoencoder framework provides a universal solution that can handle both small and large datasets effectively. The model's ability to learn robust latent representations makes it adaptable to different sample sizes, while the contribution graph method universally identifies fault sources across various operating conditions, achieving both high performance and broad adaptability.
3Reliability
If deep learning methods (AE, VAE, RBM) are used to improve monitoring performance, then the ability to handle nonlinear and non-Gaussian data improves, but the device complexity and computational requirements increase
Solution Approach 1:
The patent segments the complex deep learning task into distinct functional modules: (1) variational autoencoder for feature extraction and anomaly detection, (2) contribution graph for fault source identification, and (3) control limit calculation for decision-making. This segmentation reduces overall system complexity by making each module's function clear and manageable.
Solution Approach 2:
The patent performs preliminary action by pre-training the variational autoencoder on normal operating data to establish baseline latent space distributions and control limits. This preliminary training enables the model to quickly adapt to new data streams and reduces online computational complexity, as the heavy lifting of learning fundamental patterns is completed beforehand.
4Quantity of substance
If high-dimensional process data is monitored directly, then all variables are considered, but the complexity of analysis increases and essential features may be obscured
Solution Approach 1:
The patent applies dimensionality change by mapping high-dimensional granulation process data into a lower-dimensional latent space using the variational autoencoder. This transformation preserves essential features and relationships while reducing complexity, enabling effective monitoring and fault diagnosis without losing critical information from the original high-dimensional space.
Data Source
AI summary
A monitoring and fault diagnosis method for a continuous granulation process of drug particles based on a variational autoencoder and a contribution graph is provided. By constructing and training the variational autoencoder model, this method achieves dimensionality reduction and feature extraction of high-dimensional data, and generates latent vectors through reparameterization tricks to reconstruct the input data. This method designs monitoring statistics based on KL divergence for real-time monitoring of critical quality attributes and critical process parameters in the production process. Additionally, the contribution graph method is used to evaluate the effect of each variable on the monitoring statistics, so as to realize the accurate diagnosis and location of the fault. The monitoring and fault diagnosis method has strong adaptability and high monitoring accuracy, which can detect potential faults early, avoid the accumulation of quality problems, improve production efficiency, reduce production costs, and ensure drug quality and patient safety.