TAMMS-CMMS: AI-Supported Data Analysis for More Efficient Maintenance

The digitalization of industrial maintenance is entering a new era: integrating artificial intelligence (AI) into CMMS systems (Computerized Maintenance Management Systems) supports the development of Industry 4.0 and opens up entirely new possibilities. In our latest research project, we are developing an AI-supported extension for TAMMS that goes far beyond traditional documentation and makes a meaningful contribution to intelligent, connected maintenance systems.

Traditional Databases

Traditional relational databases, such as those successfully used in the TAMMS system, store maintenance data precisely and in a structured way. However, these systems reach their limits when it comes to recognizing semantic relationships between terms such as "replace" and "exchange", which may mean the same thing in the same context. This is where AI-supported vectorization comes in as a complementary function, enabling data to be analyzed by meaning and providing deeper insights.

Example:

• One technician writes: "V-belt replaced".
• Another technician documents: "Drive belt exchanged".

Although both descriptions may refer to the same event, there is no recognizable connection between the two text-based entries in the database. This makes it harder to identify patterns or use correlations for future maintenance planning.

The Path to Better Data: Vectorization

With our new AI-supported solution, we go one step further. Vectorizing data makes it possible to represent text and its meaning in numerical spaces. This means:

• Maintenance data is stored in a "vector space", where semantic similarities between entries can be identified mathematically.
• Terms such as "replace" and "exchange" no longer appear as isolated data points, but as related vectors.

Practical Use: Our Tests

For our experiments, we collected one month of fictional maintenance data and sensor data from three machines:

1. Vectorization of maintenance data: The generated entries were first vectorized by an AI model. These vector-based data points were then uploaded into a Qdrant database (vector database) so that semantic relationships between entries could be analyzed.
2. Sensor data processing: Sensor data was stored in InfluxDB as usual. The "assistant" has access to this data, allowing targeted queries.
3. Assistant integration: Our experimental assistant was configured to use both the Qdrant vector database and InfluxDB to answer specific questions.

How Does the Assistant Work?

• A question is analyzed and vectorized as well.
• The vectorized content is compared with the vectors stored in Qdrant.
• If necessary, the assistant creates dynamic queries for InfluxDB to retrieve additional data, such as sensor measurements.

Example: Analyzing the Assistant's Answers

The following interaction shows how the experimental assistant provides precise, context-based answers to specific questions. The assistant analyzes the requests, extracts relevant information from the databases, and returns clearly structured results. It is particularly noteworthy how well the assistant recognizes the focus of each question and responds accordingly.

1. Question (unfortunately not visible in the image):

"Show me which repairs were carried out on the CNC milling machine in March and why they were necessary."

Question Analysis:

• Focus: Repairs on the CNC milling machine in March, including reasons.
• Filters: Equipment (CNC milling machine), type (corrective), period (March).

Assistant's Answer:

• March 4: Planned maintenance without issues.
• March 11: Cooling system repair due to overheating; coolant replaced and system cleaned.
• March 19: Tool wear inspection still pending due to lack of spare parts.

Note: The assistant provides precise details about the maintenance actions and explains the causes, such as overheating.

2. Question:

"How many tasks did Hans complete in March?"

Question Analysis:

• Focus: Number and type of tasks completed by technician Hans in March.
• Filters: Technician (Hans), period (March).

Assistant's Answer:

• Hans completed two tasks in March:
  • March 23: Cleaning and filter replacement on the laser unit (corrective).
  • March 24: Quarterly hydraulic inspection on PRESS-002 (not completed on time).

Note: The assistant understands that both the number and the status of tasks may be important. It is also interesting that I only wrote Hans, yet it understood that I meant Hans Schmidt, as stored in the database.

3. Question:

"Who else worked on machine PRESS-002 in March?"

Question Analysis:

• Focus: Technicians and activities on PRESS-002 in March.
• Filters: Equipment (PRESS-002), period (March).

Assistant's Answer:

• Michael Weber worked on PRESS-002:
  • March 2: Calibration of safety sensors (corrective).
  • March 17: Repair of vibration issues (bearing replacement, corrective).
  • March 24: Quarterly hydraulic inspection (still pending).

Note: The assistant identifies technicians and specific maintenance actions, clearly assigning them to dates and categories.

Conclusion of the Analysis

The experimental assistant demonstrates a strong ability to understand requests and filter out relevant information. Each answer presents the most important details clearly and in a structured way, such as equipment, date, type of maintenance, and status (completed or pending). This underlines the effectiveness of vectorized data analysis and the system's ability to answer complex questions precisely.

Future Potential

The possibilities of AI-supported maintenance analysis are almost limitless. Potential applications include:

• Statistics: Automatic reports on frequent causes of faults.
• Documentation: Intelligent search functions in maintenance logs and machine documentation.
• Report summaries: Automatic generation of clear management reports.
• Skill matrix: Automatic evaluation of maintenance personnel's experience with specific machines.
• Inventory analysis: Analysis of stock levels with automated report generation.
• Questions: Answering specific questions about machines and personnel based on available data.
• TPM planning: Automatic creation of TPM plans based on available data and uploaded machine documents.

Conclusion

Integrating AI into CMMS systems such as TAMMS offers enormous potential for optimizing maintenance. Intelligent analysis of maintenance and sensor data helps make processes not only more efficient, but also more sustainable. With our current research work, we are laying the foundation for the next generation of maintenance digitalization.