Introduction
Maintenance, repair and operation (MRO) in Ukrainian industrial enterprises face constant pressure: the need to minimize downtime, optimize inventory costs and compensate for the shortage of qualified personnel. Traditional approaches to predictive maintenance often lead to premature replacement of components or, conversely, to unplanned stoppages due to unexpected failures. The modern solution to this problem lies in the implementation of artificial intelligence and intelligent technical assistants, which integrate data from condition sensors, enterprise resource planning systems (ERP) and technical documentation.
In the context of compliance with international and national standards, such as ISO 13374 (machine condition monitoring and diagnostics) and DSTU EN 13306 (technical maintenance terminology), the use of artificial intelligence systems enables a transition from reactive to predictive technical maintenance. Artificial intelligence does not replace experienced chief engineers, but serves as a powerful decision-making support tool, which quickly processes arrays of operating parameter data.
How It Works: Technical Basis of the System
Intelligent technical assistants are based on a combination of machine learning (ML) methods and natural language processing (NLP). The architecture of such systems solves two main tasks: analysis of time series data from vibration, temperature, and pressure sensors, as well as searching in huge technical manuals, drawings, and maintenance schedules.
To detect anomalies in the operation of pumps, compressors or reducers, algorithms use a model based on Isolation Forest or LSTM (Long Short-Term Memory) neural networks. Unlike simple threshold values (for example, stopping the pump when pressure exceeds 10 bar), deep learning models analyze complex correlations between parameters. For example, the system may detect micro changes in the bearing vibration spectrum at a constant temperature three weeks before critical wear, which corresponds to the requirements ISO 10816 regarding vibration state control.
Data requirements
The effectiveness of any artificial intelligence model directly depends on the quality of input data. Attempts to implement intelligent analytics on outdated or incomplete data inevitably lead to false alarms. For full functionality of the assistant, the following types of data are required:
- Condition Monitoring Data: Vibration sampling frequency not less than 1 kHz, temperature readings with accuracy up to 0.1 °C, current consumption data by electric motors.
- Repair History: Detailed logs of previous incidents, part replacements and fault codes according to international classifiers.
- Technical documentation: Catalogs of spare parts, operation instructions and material specifications in machine-readable formats.
Minimum volume of historical data for training the failure prediction model ranges from 6 to 12 months of continuous equipment operation. All sensors must be calibrated according to current metrological regulations.
Implementation Architecture
Typical architecture of an intelligent MRO system is built according to the hierarchical principle of operational technologies (OT) and information technologies (IT):
- Field Level: Collection of raw data from vibration sensors, pressure sensors and flow meters installed on critical equipment.
- Edge Computing: Primary noise filtering directly near the unit using industrial controllers (PLC), reducing network load.
- Cloud or On-Premise Server (Cloud/On-Premise Server): Data storage in time-series databases and execution of heavy machine learning model calculations.
- User Interface (HMI / Web App): Displaying recommendations for the repair engineer as notifications on a tablet or work station, indicating the probability of failure and a list of required spare parts.
Real results and economic effect
Implementation of intelligent technical assistants in production processes demonstrates clear quantitative efficiency indicators, confirmed by practice of leading enterprises:
| Indicator | Traditional approach | With an intelligent assistant | Change |
|---|---|---|---|
| Mean Time Between Failures (MTBF) | 4 500 hours | 7 200 hours | +60% |
| Machine Downtime | 48 hours/month | 14 hours/month | -70% |
| Cost of emergency purchases of spare parts | High | Minimum | -45% |
| Return on Investment (ROI) | N/A | 12-18 months | Positive |
Typical budget for a pilot project for one production line ranges from 25 000 to 60 000 euros, depending on the number of already installed sensors and the complexity of the technological process.
Limitations and Possible Risks
Despite technological advantages, artificial intelligence in MRO has a number of objective limitations:
Штучний інтелект не є універсальною панацеєю. Модель не здатна правильно інтерпретувати аномалію, якщо вона спричинена фізичним пошкодженням датчика або раптовою зміною технологічного рецепту без оновлення контекстних даних у системі.
The main risks include a high percentage of false positive activations at the initial stages of model training, resistance from personnel to changes in work processes, as well as cybersecurity when integrating OT networks with corporate IT systems. It is necessary to strictly adhere to the standards of industrial network security (IEC 62443).
"Build vs Buy" Approach
What is the question facing the engineering management of the company: to develop a solution internally with the internal IT department or to purchase a ready-made commercial platform?
Developing custom models requires maintaining a data scientist team and constant code support, which is economically unfeasible for medium-sized industrial companies. Commercial platforms with ready-made libraries of models for pumps, reducers and motors allow reducing the system deployment time from several years to several weeks.
Practical steps for the engineering team
For a successful launch of digital transformation of MRO on the factory, it is recommended to follow the following algorithm:
- Equipment Criticality Audit: Identify 10-20% of equipment whose failure stops the entire plant (analysis according to RCM - Reliability-Centered Maintenance).
- Data availability check: Assess the condition of existing sensors and the quality of historical data regarding repairs.
- Pilot project launch: Select one unit (for example, main exhaust fan or compressor) for installation of the test monitoring system.
- Supplier Integration: Connect reliable spare parts suppliers to automatically generate purchase orders based on wear predictions.
Conclusion
Smart technical assistants transform the role of the maintenance engineer from a mere incident responder to a strategic reliability analyst. The combination of predictive analytics with a quality technical database allows avoiding millions in losses from production downtime. For selecting compatible components, sensors and repair kits that meet European quality standards, we recommend using UNITEC-D E-Catalog.
Bibliography
- ISO 13374: Condition monitoring and diagnostics of machines — Data processing, communication and presentation.
- DSSTU EN 13306: Technical Maintenance. Terms and Definitions.
- ISO 10816: Mechanical vibration — Evaluation of machine vibration by measurements on non-rotating parts.
- IEC 62443: Security for industrial automation and control systems.
