Built to scale
Every decision about our technology stack was made with scalability in mind. The benefit for customers? An ability to rapidly deploy throughout lines and cells in a plant, and even across a global network of plants. The path from pilot to roll-out is fast.
Scalability is driven by multiple components of our proprietary stack, such as:
- Automated stack deployment
We spin up a unique hybrid on-prem / cloud architecture for each customer using state-of-the-art techniques and security practices. Deployment is fully scripted.
- High-performance data streaming platform
We developed a proprietary data streaming platform based on Apache Kafka , an open-source distributed event streaming platform used by thousands of companies for high-performance data pipelines. Our platform can be quickly and elastically expanded for increased storage and processing.
- Service-based architecture
The Liveline Digital Controls Platform (LDCP) is built around services and RESTful APIs. This allows for rapid introduction of new features and capabilities.
- Automated data selection.
Our proprietary Liveline Data Quality Score (LDQS) continually inspects incoming data and assesses its suitability for use in training models. “Good” data is automatically tagged for training without any human intervention. Typically, the LDQS can find sufficient training data simply by observing normal production without requiring the plant to perform parameter sweeps or log test pulse responses.
- Automated modeling
We use a patent-pending, 2-step method to generate AI controllers. First we create a model of your plant’s physics. This model, called a Liveline Physics Package (LPP), encodes the temporal relationships between process parameters and output metrics of interest, and can predict future output values based on current conditions. In the second step, the AI controller contained in our Liveline Controls Package (LCP) uses the LPP to learn how to manipulate process parameters and achieve optimal output. LCP performance is graded, and superior controllers are released for production. All of these steps are automated and do not require human intervention.
- Automated loading of models based on production recipes
Based on the “recipe” or combination of product SKU and production line, the appropriate LCP is automatically loaded to enable AI-based controls.
- Automated monitoring of model performance
LCP performance is continually evaluated using a Liveline Test Harness (LTH). The LTH is configured to reflect your desired performance metrics. Evaluation is automatic. Depending on the results, training for updated models can be triggered as part of our automated MLOps pipeline.
- Automated triggers for model updating or re-training
Depending on the results from the LTH, training for updated models can be triggered as part of our automated MLOps pipeline.
We do not have armies of data scientists tuning models for customers — instead, they are constantly improving our automated MLOps pipeline.
The only hands-on portion of our process is the initial mapping of data tags from your plant’s equipment to the Liveline Digital Controls Platform (LDCP), and interviews with your process experts. We incorporate their knowledge about key process variables, equipment safety limits, and other constraints, and what measurable outcomes are most important.