1. Visual Analysis

• How it Works: The system processes video and images through a Convolutional Neural Network (CNN), a form of machine learning AI. This AI is trained on extensive data from previous installations to identify and classify the presence and severity of mist and liquid.
  
• Human Verification: The AI's initial classifications are then subject to human verification in an "overview" stage, ensuring accuracy and refining the model.
• Scoring and Trends: The system scores one frame per minute, aggregating these scores every 30 minutes. This data is used to generate trend analyses, helping to determine if pipeline conditions are improving or worsening, identify process upsets, and track their duration. Heat maps are also populated to provide a quick visual overview of conditions over time.
  

2. Metadata Extraction

Metadata Extraction focuses on quantifiable features derived from the video data, with image brightness being a primary indicator.

• Image Brightness as a Key Metric: Since the LineVu camera provides its own stable light source within the pipeline (where no other light sources exist), changes in image brightness directly correlate with the amount of material reflecting that light.
• Interpreting Brightness:
  • Higher Brightness: Indicates more mist or liquid reflecting light. For instance, stratified flow (liquid pooling on the pipe floor) creates a flat surface that reflects a significant amount of light, resulting in very high brightness readings.
  • Lower Brightness: Suggests less material reflecting light, indicating clearer conditions.
• Stability Measurement: The system measures average, minimum, and maximum brightness over 60-second intervals. A wide range between min and max brightness suggests "flickering" or rapidly changing conditions, while a narrower range indicates more stable conditions.
• Scale: Brightness is measured on a scale of 0 to 255, representing the average pixel brightness across the camera's view. This data can be used to trigger immediate alarms for significant changes.

3. PhaseVu AI

PhaseVu AI represents an advanced analytical approach that integrates process data with visual observations to build a predictive model of pipeline behavior.

• Model Building: This method involves building a model that correlates your operational process data (such as flow rate, pressure, temperature, BTU, or liquid levels from knockout drums) with the expected brightness and mist levels observed by the LineVu camera.
• Predictive Power: The AI learns to predict what the camera should be seeing based on the provided process data.
• Problem Identification: The true power of PhaseVu AI emerges when the model "breaks"—meaning the camera's observations deviate significantly from the AI's predictions. This indicates an unexpected process change or problem within the pipeline, such as internal vessel damage or a filter malfunction. By identifying when the model breaks, the system can help pinpoint the root cause of issues (e.g., whether it's flow-related, temperature-related, or pressure-related).
• Data Integration: PhaseVu AI utilizes post-fact SCADA/DCS data exports, typically in CSV or Excel format. While targeted data windows are useful, providing broader datasets with varied conditions (e.g., different temperatures, flow rates) helps build a more robust model. This data can be uploaded directly through the portal.
• Value-Added Insight: This optional feature offers significant value by transforming raw data into actionable insights, helping you understand the complex interplay between process conditions and pipeline integrity.