Artificial Intelligence in Cross-Reference: Automatic Search for Equivalent Spare Parts

Introduction: MRO Challenges and AI Solutions

Modern industrial production in Ukraine faces significant difficulties in ensuring the continuity of production processes, especially in the area of maintenance and repair (MRO). One of the critical challenges is the efficient search for equivalent spare parts. This is due to the huge nomenclature, the variety of manufacturers, constant changes in the assortment and outdated catalogs.

Traditional search methods often require a significant investment of time and deep knowledge of the engineering staff, resulting in:

• Long-term equipment downtime, which reduces the overall effectiveness of the equipment (OEE).
• Purchases of excess inventory for hedging, which freezes working capital.
• High operational costs associated with logistics and warehouse maintenance.

The solution to these problems is the application of artificial intelligence (AI) for automated cross-reference of spare parts. This technology makes it possible to identify functional and technical equivalents of original components, significantly optimizing MRO processes.

Thanks to AI solutions, enterprises can achieve:

• Reduced downtime due to accelerated component sourcing and delivery.
• Optimization of warehouse stocks, minimizing redundancy and related costs.
• Reduction of total costs for the purchase and maintenance of spare parts.

Principles of AI Work for Cross-Reference

The main idea is to use machine learning (ML) algorithms for deep analysis of technical characteristics, geometric dimensions, chemical composition of materials, operational parameters and other metadata of spare parts. AI models are able to detect hidden patterns and relationships that cannot be easily identified by humans or traditional systems.

The process of cross-referencing using AI includes several key steps:

1. Data Collection and Integration:

   A variety of information is collected from databases of ERP systems, PDM systems, maintenance management systems (MMS), electronic catalogs (for example, UNITEC-D E-Catalog), operating instructions and technical drawings. It is important to integrate this data into a single, centralized system.

2. Pre-Processing and Normalization of Data:

   Raw data often contains discrepancies, omissions, or non-uniform formats. At this stage, natural language processing (NLP) methods are used to standardize textual descriptions, as well as computer vision algorithms to analyze images and drawings. For example, the description "ball bearing radial 6205" can be normalized to a format containing separate fields for type (ball), purpose (radial), and marking (6205).

3. Learning a Machine Learning Model:

   Learning algorithms, often based on neural networks or ensemble methods, analyze normalized data to build complex models. These models learn to identify the degree of similarity between different components. For example, for bearings, the AI ​​model can compare them according to parameters such as inner diameter (d), outer diameter (D), width (B) according to DSTU GOST 520:2014, EN ISO 281, load capacity (C, C₀), maximum rotation frequency, tolerances (accuracy class). The model also takes into account separator materials, lubricant type and operating temperature.

4. Formation of Recommendations:

   After training, the model can receive a request for a specific spare part and generate a list of potential equivalents. Each recommendation is accompanied by an indicator of relevance or probability of equivalence. For example, for a bearing with an inner diameter of 25 mm, an outer diameter of 52 mm, and a width of 15 mm, the model may suggest several options from different manufacturers, indicating the deviation in micrometers (μm) from the ideal fit.

Data Requirements: Volume, Quality and Format

The quality and volume of input data are critical to the effectiveness of any AI system. The rule "garbage in, garbage out" is particularly relevant here.

Types of required data:

• Technical specifications: Detailed parameters such as dimensions (mm), tolerances (μm), working pressure (bar), temperature range (Celsius), rated power (kW), rotation speed (rpm), chemical composition of materials, tensile strength (MPa). For hydraulic components, this can be throughput (l/min) or viscosity of the working fluid.
• Data about suppliers and manufacturers: Articles, serial numbers, catalog designations, information about the country of origin.
• Graphic data: Technical drawings (2D, 3D models), photographs of components. Computer vision algorithms can extract critical information from them, such as the location of the holes, the shape of the fastener, and overall dimensions.
• Operational data: Historical records of equipment failure, component lifetime (MTBF), operating conditions, procurement data. This data helps AI understand actual reliability and compatibility.

Data quality:

High quality data is the basis for accurate recommendations. Vague, inconsistent, or incomplete data lead to significant errors. This requires the implementation of data cleaning, verification and standardization processes. For example, a single format for pressure measurements (always in bars, not psi or Pa). All units must be metric.

Data volume:

The more quality data available to train the model, the more accurate and reliable its predictions will be. Achieving high accuracy often requires thousands or millions of spare part records.

Data format:

Structured data in relational databases or JSON/XML formats is preferred. However, AI systems must also work effectively with unstructured data, such as text descriptions, PDF files, images and CAD files, using specialized modules to extract and process them.

Architecture of Cross-Reference System Implementation

The effective implementation of an AI solution for cross-referencing spare parts requires a well-thought-out architecture covering data collection, data processing, analysis and integration with existing enterprise systems. A typical architecture includes the following components:

1. Data Sources:

   The basis is the company's existing systems: ERP (Enterprise Resource Planning), PDM (Product Data Management), MES (Manufacturing Execution System), warehouse management systems (WMS), as well as electronic catalogs, such as UNITEC-D E-Catalog. These systems contain detailed data about the components, their characteristics and usage history.

2. Data Ingestion Layer:

   Responsible for collecting data from various sources. Includes data adapters and APIs to communicate between systems. ETL (Extract, Transform, Load) processes are used to extract, transform and load data into a single storage (Data Lake or Data Warehouse).

3. Data Processing Platform (Data Processing Platform):

   It can be implemented both on cloud services (for example, AWS, Azure, Google Cloud Platform) and on local servers (on-premise). This platform provides the necessary computing resources for:

   • Cleanup and Normalization: Automated tools for removing duplicates, correcting errors, standardizing units of measurement, and formatting text descriptions.
   • Data enrichment: Adding additional information, for example, quality standards (ISO, DSTU), certifications (CE, UkrSEPRO).
   • Construction of Vector Representations: Transformation of complex technical characteristics and unstructured data (text, images) into numerical vectors that can be processed by ML-algorithms.

4. Modeling and Analysis Layer (ML Model Layer):

   Trained machine learning models are hosted and executed here. AI models are continuously trained on new data, improving their accuracy. The main tasks of this layer:

   • Determination of similarity between components.
   • Generating recommendations for equivalents.
   • Assessment of credibility and risks of substitution.

5. User Interface and Integration with Related Systems:

   AI results are integrated directly into user workflows. It can be:

   • Web interface for engineers and procurement professionals.
   • Integration with existing ERP systems, procurement systems or EMS, allowing you to receive recommendations directly in familiar work tools.
   • API for other programs to access the cross-reference functionality.

6. Security and Monitoring Layer:

   Ensures the protection of confidential data in accordance with the requirements of ISO 27001. Includes systems for monitoring the performance of AI models and their accuracy, as well as managing user access.

This architecture allows you to create a reliable, scalable and secure system that automates the complex process of finding equivalent spare parts.

Real Results and Economic Efficiency

The implementation of AI-driven cross-reference systems demonstrates significant improvements in operational performance and significant economic impact in industrial enterprises. Below are real-world metrics and examples:

1. Decreasing Equipment Downtime:

   A machine-building plant in the Dnipropetrovsk region, Ukraine, after implementing an AI system that allowed to quickly find analogues of critical components, recorded a reduction in downtime of production lines by 15%. This was achieved thanks to the accelerated search for spare parts, which used to take up to 4-8 hours, but now takes 15-30 minutes.

2. Optimization of Warehouse Stocks:

   At one of the metallurgical plants of Ukraine, the AI ​​system helped to identify duplicate or redundant positions of spare parts, which led to a reduction of warehouse stocks by 20%. This freed up around €500,000 worth of working capital, which was reinvested in equipment upgrades. The payback period (ROI) of such a solution was approximately 12-18 months.

3. Reduction of Purchase Costs:

   The chemical industry enterprise was able to reduce the cost of purchasing spare parts to 8%. The AI ​​system automatically identified cheaper, but fully compatible and certified alternatives from reliable suppliers, while maintaining compliance with all technical standards (for example DSTU EN ISO 12100 for machine safety).

4. Increasing the Technical Readiness Ratio (TSR):

   On the production line with critical equipment, KTG increased from 0.92 to 0.95. This means that the equipment was available for work 3% more time, which directly affected the volume of output.

5. Reduction of Time for Engineering Processes:

   Instead of spending hours manually searching through paper catalogs or spreadsheets, engineers can now get accurate recommendations in minutes. This allows them to focus on more complex tasks, such as predictive maintenance or optimizing production processes.

The total cost-effectiveness of such systems often exceeds the initial investment, confirming their value to industrial enterprises focused on efficiency and competitiveness.

Limitations and Potential Risks

Although AI cross-reference systems offer significant advantages, it is important to objectively assess their limitations and potential risks. AI is a powerful tool, but it is not a universal panacea.

1. Dependence on the Quality of Input Data ("Garbage in, Garbage Out"):

   The accuracy and reliability of AI recommendations directly depend on the quality and completeness of the data it was trained on. If the input data contains errors, inaccuracies, omissions or contradictions, the system will generate incorrect or unhelpful results. This requires significant investment in data collection, cleaning and verification in the initial stages of implementation.

2. Difficulty of Working with "Non-standard" Details:

   For unique, heavily modified, or very old components for which sufficient data is lacking, AI may struggle to find equivalents. In such cases, as before, an expert assessment by experienced engineers will be required.

3. The need for Human Verification and Validation:

   The AI ​​system is an auxiliary tool that optimizes the search process. However, the final decision to replace critical components should always be left to a qualified engineer. This especially applies to parts that affect operational safety and require UkrSEPRO certification or compliance with CE directives. The engineer must check all parameters, including materials, tolerances and compatibility with the system.

4. Initial Implementation and Support Costs:

   Deploying a full-fledged AI system requires significant initial investment in software, hardware, and most importantly, skilled personnel (data processing engineers, machine learning specialists). It also requires ongoing costs to maintain, update, and retrain AI models.

5. Integration Challenges:

   The integration of a new AI system with the existing IT infrastructure of the enterprise (ERP, PDM, SUTO) can be complex and require significant efforts to ensure data compatibility and seamless exchange of information.

Given these limitations, it is important to approach AI implementation with realistic expectations and a clear understanding of the human role in the process.

Own Development or Ready Solution ("Build vs Buy")

When considering the implementation of an AI system for the cross-reference of spare parts, enterprises face a dilemma: develop their own solution (build) or purchase a commercial product (buy).

Own Development (Build):

• Advantages:
  • Full Control: Ability to create a system perfectly adapted to the unique needs, processes and IT infrastructure of the enterprise.
  • Deep Integration: Ensuring the deepest possible integration with existing internal systems.
  • Intellectual Property: The Company owns all developed technology and data.
• Disadvantages:
  • High Costs: Significant capital and operating costs for development, testing, implementation and support.
  • Time to Implementation: The development process can take 18-24 months or more, from idea to full operation.
  • Need for Specialists: The need to have on staff highly qualified specialists in machine learning, data processing, software development and domain expertise.
  • Project Risks: High risks of project failure, exceeding the budget and deadlines.

Ready Commercial Solution (Buy):

• Advantages:
  • Fast Implementation: Ability to deploy the system relatively quickly, usually in 3-6 months.
  • Proven Functionality: Getting a ready-made, tested and verified solution with already built-in functions.
  • Reduced Risks: Lower development risks because the product already exists and is supported by the vendor.
  • Professional Support: Access to expert support, updates and new features from the vendor.
• Disadvantages:
  • Less Flexibility: Limited possibilities for customization to very specific requirements.
  • Supplier Dependency:Supplier dependency in terms of functionality, pricing and development roadmap.
  • Cost of Licenses: Ongoing license and subscription costs.

UNITEC-D GmbH, as a global supplier of MRO components, understands these challenges. We offer not only high-quality certified spare parts, but also expert support in the selection and adaptation of commercial AI solutions. Our technical expertise and in-depth knowledge of spare parts nomenclature make it easy to integrate data from our E-Catalog into your AI system, regardless of the chosen strategy.

Practical Steps for Implementation

A step-by-step approach is recommended for engineering teams and production managers considering implementing an AI cross-reference solution:

1. Data Audit and Evaluation:

   Conduct a detailed audit of available spare parts data. Evaluate their completeness, accuracy and format. Identify gaps and sources of potential improvement. This is the foundation for any AI initiative.

2. Definition of the Pilot Project:

   Instead of trying to implement a system for the entire volume of spare parts, choose a small but critical group of parts (for example, bearings for a key unit or valves for a particular pipeline). This will allow you to quickly get results and evaluate the effectiveness of the solution with minimal risks.

3. Choice of Platform and Partner:

   Determine whether a cloud solution or an on-premises deployment is right for your business. Consider commercial offers from AI solution providers. UNITEC-D GmbH can provide consulting support in the selection of compatible solutions and help adapt data from our E-Catalog for integration into your system.

4. Infrastructure preparation:

   Provide the necessary computing power and data storage. Set up integration mechanisms with existing ERP and SUTO systems.

5. Training and Adaptation of Personnel:

   Organize training for engineers, procurement specialists and service personnel. It is important that they understand the principles of AI and be able to use the new system effectively. Encourage feedback for continuous improvement.

6. Continuous Monitoring and Improvement:

   AI systems need regular monitoring of their performance and accuracy. Collect feedback from users and use it to retrain models, add new features, and expand the reach of the spare parts catalog.

By following these steps, Ukrainian industrial enterprises can effectively implement AI technologies to optimize cross-reference processes and increase the overall efficiency of MRO.

Conclusion

The use of artificial intelligence for automated cross-reference of spare parts is a significant step in the direction of digitalization of industrial MRO. This technology allows Ukrainian enterprises to significantly improve the efficiency of inventory management, reduce equipment downtime and reduce operating costs. AI systems, analyzing huge amounts of technical data, provide accurate and fast recommendations, allowing engineers to make informed decisions about replacing components.

Despite certain limitations and the need for investments in data and infrastructure, the economic return from the implementation of AI solutions is significant. UNITEC-D GmbH, as a reliable partner in the field of industrial components, is ready to provide high-quality certified spare parts according to DSTU, EN and ISO standards, as well as support your team in implementing advanced MRO technologies.

To obtain high-quality certified spare parts and optimize M&R processes, visit UNITEC-D E-Catalog.

Link

• DSTU GOST 520:2014 Rolling bearings. General technical conditions.
• EN ISO 281:2007 (DSTU ISO 281:2018) Rolling bearings. Dynamic and static nominal load capacity.
• ISO 27001:2022 Information technologies. Methods and means of ensuring security. Information security management systems. Requirements
• Directive 2006/42/EC on machinery (CE marking).
• Technical regulation on machine safety (UkrSEPRO).
• DSTU EN ISO 12100:2016 Machine safety. General design principles. Risk assessment and risk mitigation.