Oil Pump Failure Prediction Using Federated Multi-Label Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive maintenance models for oil pump maintenance in the oil and mining industries face limitations due to the lack of data sharing between sites, privacy concerns, and the need for site-specific models, which hampers the development of accurate predictive models for asset failure.
Innovation Solution
A distributed learning approach is implemented, where assets are partitioned into static, semi-static, and dynamic features, forming cohorts, and local models are generated and shared to create a global model, allowing for predictive failure analysis without compromising privacy. This involves generating local models at each site, pooling them centrally, and choosing the best model for each cohort to predict asset failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is shared between sites to improve model accuracy, then predictive model accuracy is improved, but privacy and security are compromised
Solution Approach 1:
The system segments the centralized model training process into distributed local model training at each site. Each site trains its own model locally using its data, then only model parameters (not raw data) are shared and aggregated. This segmentation allows multiple sites to contribute to a global model without exposing their sensitive operational data, thus improving predictive accuracy while maintaining privacy and security.
Solution Approach 2:
The patent introduces a federated learning intermediary layer that acts as a mediator between local sites and the central aggregation system. This intermediary manages the exchange of model parameters while preventing direct access to local data, serving as a trusted intermediary that enables collaboration without compromising privacy or security of the underlying operational data.
2Ease of operation
If a universal predictive model is built to simplify deployment, then ease of operation is improved, but adaptability to local conditions deteriorates
Solution Approach 1:
The model system is segmented into a global model component and local model components. The global model provides a standardized baseline that ensures ease of deployment across all sites, while local models are trained site-specifically to adapt to local conditions. This segmentation allows the system to maintain both operational simplicity through the global model and local adaptability through customized local models.
Solution Approach 2:
The system implements dynamic model selection where the predictive model can switch between the global model and local models based on the specific site and cohort. This dynamic approach allows the system to automatically select the most appropriate model for each local context, balancing the ease of operation from a universal model with the adaptability of site-specific models.
3Object-affected harmful factors
If local models are trained independently at each site to maintain data privacy, then privacy is preserved, but model accuracy deteriorates due to lack of data sharing
Solution Approach 1:
The system merges multiple local models into a global model through federated averaging of model parameters. Each site trains its local model independently on its own data (preserving privacy), then the model parameters are aggregated to create a global model that incorporates knowledge from all sites. This merging process improves predictive accuracy by combining diverse local insights without requiring raw data sharing.
Solution Approach 2:
The federated learning system implements a feedback loop where local models are trained, evaluated, and their parameters are aggregated into the global model, which is then redistributed to all sites for further local training. This iterative feedback process continuously improves model accuracy while maintaining data privacy, as each iteration refines the global model based on local performance without exposing sensitive data.
Data Source
AI summary
A computer implemented federated learning method of predicting failure of assets includes generating a local model at a local site for each of the cohorts and training the local model on local data for each of the cohorts for each failure type. The local model is shared with a central database. A global model is created based on an aggregation of a plurality of the local models from a plurality of the local sites. At each of the plurality of local sites, one of the global model and the local model is chosen for each of the cohorts. The chosen model operates on local data to predict failure of the assets. The utilized features include partitioning features of the assets into static features, semi-static features, and dynamic features, and forming cohorts of the assets based on the static features and the semi-static features.