add.AI Predictive Maintenance Model (Proof of Concept)

Predictive maintenance solution is part of the add.AI solution portfolio which is leveraging power of Artificial intelligence to improve operation and business.

Key Benefits of the solutions:

• Reduce downtime with predicting planned and unplanned stops and improve productivity.
• Understand reasons for the machine failures.
• Alerting and notifications: make corrective actions on machines and/or product line upfront before they stop working.
• Get insights on your asset’s health.
• Empower your employees to take smart decisions.

How it works:
Predictive Maintenance solution is built with the Machine Learning technology to learn from historical data and real time data to analyse failure patterns. Since conservative procedures result in resource wastage, Predictive Maintenance using Machine Learning looks for optimum resource utilization and predicting failure before they happen, on the machine level (predictive maintenance) and on the product level (predictive scrap).

About solution:
To develop relevant Machine learning models, two types of data sources are required:

• Data from Machines and Sensors: all data from sensors that monitor machinery parameters and behaviour (all data generated on the machines such as temperature, pressure, raw material quality, etc.), including data about errors. Data will be processed in real-time.
• Structured Data: structured data available for the production such as work orders, production status, production planning, material stock, etc.

• Azure IoT Hub: To provide a cloud-hosted solution back end to connect virtually to any device. Service is needed for capturing real time data. It offers possibility of registration machines and/or sensors in production in cloud as individual devices (eg. A machine or several machines are registered as a Device in Azure IoT Hub serves as client for data transmissions generated on the machines).
• Azure Stream Analytics: To provide real time parsing of stream data (transformation, aggregation of data). Parsed data are input for:
• Power BI real time analytics and alerting (presentation layer and analytics).
• Azure Data Lake for long term storage of data (cold storage of data).
• Azure Synapse for dimensional models’ creation and reporting for non-real time data which can enriched with other data from production. Presentation layer in Power BI.
• Azure Databricks: For predictive models learning and near real-time forecasting. Proposed solution anticipates forecasting in different time intervals (real-time and/or 5 minutes and/or 30 minutes and/or 1 hour).
• Supporting services (Azure AD, Azure Key Vault, Azure Logic Apps, Azure Functions, Azure DevOps, Azure Monitor) are used to ensure security, orchestration, and monitoring of Azure services utilization.

Timeline- Agenda:

• 1st - 2nd week: Envisioning Workshops to identify data sources required for AI model. GAP-FIT Assessment
• 3rd - 6th week: Data integration and deployment of AI model (CRISP methodology)
• 6th - 8th week: Final revision with the customer, Model & Results presentation, next steps proposals

Payment model: SaaS

• Monthly Fee = FIX (predictive maintenance solution fee * number of machines) + FLEX (Azure Services).
• Price and Scope will vary depending on the requirements, size, and complexity of your environment.