Manufacturing SMEs: when the smart factory and AI move from theory to reality

In our previous blog post, "Manufacturing SMEs: Smart Factory and AI, a Winning Duo," we established that data reliability is the essential foundation for any artificial intelligence initiative. Indeed, systems must "talk" to each other to transform raw data into strategic intelligence.

In this blog post, we move to the next step. We explore three concrete use cases inspired by our achievements: OT (Operational Technology) data architecture: the foundation of manufacturing AI projects, dynamic allocation of carpet production and operators, and customer quote assistance.

OT Data Architecture: The Foundation of Manufacturing AI Projects

In many factories, AI projects fail not because of models or algorithms, but because PLC, SCADA, MES, and other system data are not structured correctly. Without an OT foundation, AI cannot transform this data into reliable and actionable decisions.

Challenge: Difficult-to-Use Data

In most factories, shop floor data comes from a stack of systems (e.g., PLC, HMI, SCADA, MES, quality systems, CMMS) installed at different times. These systems work, but they were never designed to feed advanced analytics or AI models. The result: a lot of data is accumulated, but very little is actually usable.

• Scattered data: Information is distributed across multiple systems without a common data model.
• Non-standardized naming: Each vendor or integrator uses their own tag conventions.
• Lack of context: Data is rarely linked to equipment, products, batches, recipes, or shifts.
• Inadequate historization: Data is either missing or stored at a granularity unsuitable for advanced analysis.
• OT–IT gap: It is difficult to connect production data to business systems like ERP, quality, or inventory.

Solution: Create a Reliable and Consistent OT Architecture

To make shop floor data usable by AI, you must first create a structured and consistent OT architecture. The goal is to centralize, standardize, and contextualize data at the source so it can feed advanced analytics and predictive models.

• Data centralization: Gather data from PLC, SCADA, MES, and other systems into a single platform.
• Tag standardization: Harmonize naming conventions and units to facilitate information comparison.
• Contextualization: Add information about equipment, product, batch, recipe, or shift to enrich analyses.
• Adapted historization: Store data with the granularity needed for different types of analysis and AI models.
• OT–IT integration: Progressively connect production data to ERP, quality, and maintenance systems for a complete and consistent view.

Benefit: Confidently Leverage Industrial AI with Organized Data

Once the OT foundation is in place, AI produces more value. Models leverage reliable, consistent, and contextualized data, facilitating decision-making and factory-wide deployment.

• More reliable analyses: Models and reports are based on consistent and reproducible data.
• Time savings: Less time spent cleaning or correcting data to make it usable.
• Team confidence: Operators, managers, and engineers can trust the generated recommendations.
• Scalable deployment: Foundation models can be applied to multiple lines or factories.
• Return on investment: AI projects deliver value faster and measurably.

Dynamic Allocation of Carpet Production and Operators

This use case illustrates how AI solves the puzzle of carpet production planning and scheduling, optimizing equipment and workforce utilization.

Challenge: Bottlenecks and Setup Change Costs

Carpet manufacturing involves a complex sequential flow (Extrusion → Tufting → Oven → Finishing) often distributed across multiple distinct production units. This challenge is exacerbated by tangible factors and constraints that paralyze manual planning:

• Flow complexity and problem size: The production flow is complex with multiple constraints. With infinite possible production schedule combinations, the planner struggles to satisfy all constraints and optimize production.
• Critical bottleneck (the oven): The oven is the main bottleneck, requiring long setup times to switch from one product type to another.
• Extremely costly setup times (tufting): The tufting stage requires tedious spool changes, making setup time minimization critical for profitability.
• Resource micromanagement: Static operator allocation between manual finishing stations and automated cells causes inefficiency and time losses.
• Inadequate software: Traditional planning and scheduling software (ERP/MES/APS) often tries to solve all planning and scheduling problem variants. Contrary to popular belief, this makes them unsuitable for real industrial challenges.

Static assignment and manual planning are inadequate for managing these complex interdependencies.

Solution: AI at the Heart of Scheduling

AI addresses this challenge by building a global vision and using a custom approach integrating multiple combinatorial optimization techniques (e.g., constraint programming, large neighbourhood search) to find the optimal production sequence:

• Global orchestration: AI intelligently explores combinations that minimize total production and setup time. It favours campaigns (multiple carpets of the same type) in Tufting to reduce non-value-added setup times.
• Feasibility guarantee: AI guarantees a valid solution that respects the production flow and all operational constraints while minimizing setup times.
• Dynamic allocation and real-time data: The system integrates industrial data in real time to generate a continuous plan optimizing equipment use and operator tasks (e.g., switching between manual cutting and automated finishing).
• Targeted approach: The goal is to use AI to solve complex, targeted problems, then combine solutions. For example, an optimized production schedule greatly facilitates operator assignment across machines and schedule management optimization.

Benefit: Hidden Cost Reduction and Throughput Increase

This approach enables greater agility, a decisive advantage:

• Custom approach: Unlike generic APS, this solution is precisely adapted to the real industrial challenge and specific carpet manufacturing constraints. It generates significantly superior gains.
• Capacity maximization: Significant production throughput increase by maximizing Oven operating time (critical bottleneck) and reducing non-productive downtime.
• Changeover cost reduction: Significant operating cost reduction by minimizing long and costly setups, directly translating to better margins.
• Efficient personnel management: Major facilitation of assignment, schedule management, and workforce planning through better production predictability.
• Agility facing the unexpected: AI generates and adapts complex schedules in minutes, enabling response to new orders or unexpected breakdowns.

Customer Quote Assistance

Challenge

For manufacturing SMEs, preparing a customer quote is a long, complex, and meticulous process. Each request imposes precise technical requirements, material specifications, and analysis of various drawings (e.g., AutoCAD, PDF). This process faces several major obstacles:

• Intensive manual research and information fragmentation: Collecting necessary information (historical quotes, reference manuals) is extremely fragmented. This requires manual examination of dozens or even hundreds of documents, consuming valuable time.
• Dependence on critical human expertise: Correctly interpreting requirements, evaluating risks, and writing an accurate quote requires considerable expertise. This dependence makes the process vulnerable.
• Strategic knowledge loss: With key employee departures, expertise and critical data on past projects are often lost. This knowledge drain affects future estimate accuracy and the ability to replicate past successes.

These combined challenges significantly slow the sales cycle and limit the volume of quotes the team can handle, thus hindering growth.

Solution: RAG Artificial Intelligence

AI, particularly RAG approaches, transforms this process by leveraging the company's documentary heritage. Specifically, it aims to:

• Universal data access: The RAG system quickly identifies and structures historical quotes, profitable projects, and technical documents, regardless of their original format (e.g., PDF, JPEG).
• Targeted recommendations: AI then uses these elements to automatically generate a detailed quote draft and perform a forward-looking profitability analysis.

Benefit

The main advantage is a drastic acceleration of quote generation and manual document search. This approach also enables:

• Reduced preparation time and ability to respond to a higher volume of requests.
• Improved estimate accuracy based on exhaustive data.
• Standardized format and quote quality.
• Increased contract win rate and maximized targeting of relevant offers.
• Leverage of past projects and knowledge retention.