Building Analytics Apps

Tier0 adopts Marimo Notebook to do advanced data analysis with Python, and a Bowtie application is used as an example to demonstrate the process.

flowchart LR
  collect["Node-RED"] -->|"collected data"| uns[("UNS<br/>modeling")]
  uns -->|"raw data"| notebook["Notebook"]
  notebook -->|"analysis results"| uns
  uns -->|"analysis results"| builder["App Builder"]

  classDef t0accent fill:#EAF8C8,stroke:#A6CF38,stroke-width:1px,color:#171717
  classDef t0soft fill:#F7FAF2,stroke:#D8E6B8,stroke-width:1px,color:#2A2A2A
  classDef t0agent fill:#EEF4FF,stroke:#B7C7E8,stroke-width:1px,color:#1F2937
  class uns t0accent
  class collect,notebook t0soft
  class builder t0agent

Example Background

Refinery corrosion is a complex process influenced by multiple operating factors. This example demonstrates a real-time corrosion risk assessment workflow that combines process data to estimate the likelihood of corrosion and support proactive maintenance.

Getting Raw Data

1. In Tier0, go to UNS, and import the following models.

   • Corrosion Monitoring

     {
       "name": "Corrosion_Monitoring",
       "topic": "Aramco/CDU_Plant/Atmospheric_Overhead/Metric/Corrosion_Monitoring",
       "type": "Metric",
       "description": "Real-time process measurements for CDU atmospheric overhead corrosion monitoring",
       "fields": [
         {
           "name": "D103_CHLORIDE",
           "dataType": "FLOAT",
           "unit": "ppm",
           "description": "D-103 Overhead Reflux Drum Chloride concentration"
         },
         {
           "name": "D103_PH",
           "dataType": "FLOAT",
           "unit": "pH",
           "description": "D-103 Overhead Reflux Drum pH value"
         },
         {
           "name": "WASH_WATER_FLOW",
           "dataType": "FLOAT",
           "unit": "t/h",
           "description": "Overhead wash water flow"
         },
         {
           "name": "DESALTER_SALT_PTB",
           "dataType": "FLOAT",
           "unit": "PTB",
           "description": "Salt content in desalted crude"
         },
         {
           "name": "DESALTER_BSW",
           "dataType": "FLOAT",
           "unit": "%",
           "description": "Basic sediment and water content"
         },
         {
           "name": "WASH_WATER_RATE",
           "dataType": "FLOAT",
           "unit": "%",
           "description": "Wash water rate"
         },
         {
           "name": "RRD_PH",
           "dataType": "FLOAT",
           "unit": "pH",
           "description": "Reflux drum sour water pH"
         },
         {
           "name": "RRD_CHLORIDE",
           "dataType": "FLOAT",
           "unit": "ppm",
           "description": "Reflux drum chloride concentration"
         },
         {
           "name": "RRD_TOTAL_IRON",
           "dataType": "FLOAT",
           "unit": "ppm",
           "description": "Total iron concentration indicating corrosion"
         }
       ]
     }

   • Corrosion Risk

     {
       "name": "Corrosion_Risk",
       "topic": "Aramco/CDU_Plant/Atmospheric_Overhead/State/Corrosion_Risk",
       "type": "State",
       "description": "Bayesian Network inferred corrosion risk state",
       "fields": [
         {
           "name": "risk_state",
           "dataType": "STRING",
           "description": "Current corrosion risk state: NORMAL, DEVELOPING, CONFIRMED"
         },
         {
           "name": "previous_state",
           "dataType": "STRING",
           "description": "Previous corrosion risk state"
         },
         {
           "name": "confidence",
           "dataType": "FLOAT",
           "unit": "%",
           "description": "Inference confidence"
         },
         {
           "name": "timestamp",
           "dataType": "DATETIME"
         }
       ]
     }

   • Corrosion Risk Probability

     {
       "name": "Corrosion_Risk_Probability",
       "topic": "Aramco/CDU_Plant/Atmospheric_Overhead/Metric/Corrosion_Risk_Probability",
       "type": "Metric",
       "description": "Bayesian Network posterior probability results",
       "fields": [
         {
           "name": "P_NORMAL",
           "dataType": "FLOAT"
         },
         {
           "name": "P_DEVELOPING",
           "dataType": "FLOAT"
         },
         {
           "name": "P_CONFIRMED",
           "dataType": "FLOAT"
         },
         {
           "name": "LOPC_PROBABILITY",
           "dataType": "FLOAT"
         },
         {
           "name": "SHUTDOWN_PROBABILITY",
           "dataType": "FLOAT"
         },
         {
           "name": "ESCALATION_PROBABILITY",
           "dataType": "FLOAT"
         }
       ]
     }

2. Go to Flows, create Source Flow to connect raw data and publish to UNS.

   (Node-RED connects data and publishes it to UNS.)

Building Analytic App in Notebook

1. In Tier0, go to Notebook, and create a new notebook.

2. Access the notebook, and add the following cells to analyze data from UNS.

   • Cell 1: Get data from UNS

   • Cell 2: Validate and organize data

   • Cell 3: Discretize continuous values

   • Cell 4: Define Bayesian network structure and CPTs

   • Cell 5: Build the network and run inference

   • Cell 6: Organize analysis results

   • Cell 7: Write results back to UNS through MQTT

3. Run all cells and go to UNS to check the results.

Building Bow-tie App

1. In Tier0, go to Builder.

2. Enter the application requirements in the dialog, and start building.

   What must be stated

   Remember to tell the agent to use data from UNS, specific topic paths are necessary. Otherwise the agent might have trouble displaying the correct data.

   prompt

3. Once the application is complete after certain rounds of refining, click Deploy at the upper-right corner.

4. Go to Launchpad, open the application and check.