Data Analytics & AI for Engineers | Altair RapidMiner

Roland Jones, a solution specialist at Altair, provides a comprehensive walkthrough of the Altair RapidMiner platform, demonstrating how it can be used to build a complete predictive maintenance solution from the ground up. The presentation showcases the platform’s capabilities, moving from raw, siloed data to a sophisticated, interactive AI agent for operator support.

Introduction to Altair RapidMiner

Altair RapidMiner is presented as an end-to-end data analytics and AI platform designed to bridge the gap between disparate data sources (Excel, databases, cloud platforms) and actionable business insights. The platform is built on two core layers:

1. A Connected Data Fabric: This layer acts as a universal data overlay, leveraging knowledge graphs to connect structured and unstructured data from various sources in a vendor-neutral way.
2. An AI and Automation Toolset: This is a comprehensive suite designed for a diverse range of users, from code-savvy data scientists to domain experts like engineers who may not have a coding background. Key features include:
   • For Coders: Integration with VS Code and Jupyter notebooks, along with centrally managed Python environments.
   • Low-Code/No-Code Workflows: A visual, block-based environment (Workflow Studio) for building data preparation and predictive analytic pipelines without writing code.
   • Generative AI: Tools like Prompt Studio to integrate LLMs with production data and Agent Studio to build AI agents.
   • AutoAI: An automated machine learning capability for quickly generating predictive models and insights from a dataset.
   • MLOps: Functionality to deploy, operationalize, and monitor the performance of machine learning models in production.
   • Collaboration: The platform is built with teamwork in mind, allowing users to share, reuse, and audit processes.

Use Case: Predictive Maintenance for Robot Arms

The demonstration focuses on a real-world predictive maintenance scenario. The objective is to move from reactive or time-based maintenance to a more efficient, data-driven strategy by predicting which robot arms are likely to fail. The project is broken down into three main tasks:

1. Data Unification: Bringing together and preparing data from multiple sources.
2. Predictive Modeling: Using historical data to predict future machine failures.
3. Operator Support: Creating an AI-powered assistant to help engineers on the ground.

Phase 1: Data Preparation and Visualization

This phase is conducted within the Workflow Studio, highlighting the low-code approach.

• Data Ingestion & Cleaning: A visual workflow is constructed by dragging and dropping operators. The process starts by reading a CSV file containing robot arm specifications. Using the built-in statistics view, the data is inspected, and issues are identified, such as incorrect data types and missing values. The workflow is built to:
  • Convert a date column from a categorical type (polynominal) to a proper date format.
  • Fill in missing values for the ‘plant’ location, a step shown using the GenAI Copilot feature which interprets natural language commands like “remove missing values.”
  • Rename a column, again using the Copilot.
• Data Integration: The workflow is expanded to read data from a PostgreSQL database containing usage and failure history (breakdown status). A Join operator is used to merge the two datasets on robot ID, creating a unified and enriched dataset.
• Dashboarding with Panopticon: The cleaned data is sent to Panopticon, Altair’s visualization tool. A dashboard is quickly assembled to visualize the data, featuring:
  • A data table with conditional coloring to highlight rows where a breakdown occurred.
  • Interactive filters for factory and robot type.
  • A pie chart showing the overall proportion of breakdowns.
  • A bar chart analyzing breakdown rates per robot type.

Phase 2: Predicting Failures with AutoAI

With the data prepared, the focus shifts to predictive analytics using the AutoAI feature.

• Automated Model Building: The user selects the target column to predict (breakdown). AutoAI automatically analyzes input features, suggesting which ones to include or exclude based on stability and correlation.
• Model Training and Evaluation: Using a ‘Quick Prototyping’ setting, AutoAI trains several models (e.g., Naive Bayes, Logistic Regression, Decision Tree) and presents them on a leaderboard ranked by performance metrics like accuracy and precision.
• What-If Simulation: The platform includes a Simulator that allows users to perform what-if analysis on a trained model. By tweaking input variables like ‘hours in service’ or ‘payload,’ users can interactively see how changes impact the model’s prediction, helping them understand the thresholds for failure.

Phase 3: Text Analytics & Generative AI

This section demonstrates how to extract value from unstructured text data, such as maintenance notes.

• Traditional Text Mining: A workflow is built to process the free-text ’notes’ column. This involves standard text-processing steps like tokenization, case transformation, and stop-word removal. Topic modeling is then applied to automatically discover latent themes within the failure notes, identifying clusters of words related to specific issues like ‘joint leaks’ or ‘gear failures’.
• Generative AI with Prompt Studio: Prompt Studio is used to apply LLMs to the text data. A prompt is engineered to classify each maintenance note into one of three categories: ‘joint failure’, ‘gear failure’, or ‘other’. The key step is showing how this prompt can be productionized into a reusable workflow, which is then applied to the entire database of notes, automatically enriching the data with structured failure categories.

Phase 4: Building an AI Agent for Operator Support

The final part of the presentation culminates in building an AI agent using Agent Studio.

• Agent Architecture: An AI agent is defined as an LLM augmented with a set of tools. The process involves:
  1. Creating a new flow and adding an Agent node powered by an LLM.
  2. Equipping the agent with tools for specific capabilities. This includes built-in tools like a calculator and custom tools.
  3. A custom tool is created by providing the agent with a robot arm repair manual (a Word document). The agent can now access and reason over the content of this document.
  4. Another tool is added by connecting the agent to the previously built RapidMiner workflow that fetches the latest machine failure. The agent is given a natural language description of what the tool does.
• Interactive Demonstration: The power of the agentic system is shown with a final query: “Tell me how to fix the latest robot arm failure.” The agent intelligently executes a chain of actions:
  1. First, it calls the workflow tool to retrieve the data for the most recent failure.
  2. Then, it uses the context from that data (e.g., failure reason).
  3. Finally, it consults the repair manual document to provide the specific, step-by-step repair instructions for that particular failure.

This demonstrates a powerful, context-aware assistant built entirely within the Altair RapidMiner ecosystem, combining data analysis, document retrieval, and generative AI to provide tangible operational support.

• Tags:
• AI Agents
• Generative AI
• MLOps
• Predictive Maintenance
• Altair RapidMiner
• Low-Code AI
• AutoML
• Text Analytics