Halconia.
Let’s talk
Back to blog

Industrial AI Implementation: A Practical Guide for Production Plants

According to Halconia's method, architecture comes before the agent: we understand the operations before implementing any automation, and the main risk is…

Andrés Naves ·

Industrial AI Implementation: A Practical Guide for Production Plants

According to Halconia's method, architecture comes before the agent: we understand the operations before implementing any automation, and the main risk is starting with the tool before understanding the business. With that diagnosis, the next steps are to design, put into production with continuous assistance, and evolve alongside the client.

What practical steps should I follow in my plant to implement industrial AI, from diagnosis to production deployment?

Nuestro método tiene cuatro pasos continuos, y el orden importa: la arquitectura antes que el agente. Primero analizamos la empresa y sus procesos: qué decisiones se repiten, dónde reside el conocimiento, qué datos faltan y dónde se producen las esperas.

La capa de procesos en uso describe cómo fluye realmente una oferta, un pedido, una incidencia o una devolución, quién interviene, qué sistemas toca y dónde se atasca. Lo hacemos con observación y entrevistas ligeras, y el mapa sirve tanto como documentación operativa como de plano para construir las automatizaciones.

El método de Halconia incluye también los pasos de diseñar, donde detectamos oportunidades y diseñamos soluciones a medida, implantar, donde ponemos la solución en producción con asistencia continua, y evolucionar, donde crecemos junto al cliente de forma permanente. Según el Reglamento (UE) 2023/1230 de máquinas, un producto conforme con normas armonizadas, cuyas referencias se hayan publicado en el Diario Oficial de la Unión Europea, se presume conforme con los requisitos esenciales de salud y seguridad del anexo III que estén regulados por esas normas o partes de ellas.

What machinery requirements should I check when integrating an AI system into a production line?

Si el sistema de IA forma parte de una máquina, el Reglamento (UE) 2023/1230 de máquinas es el marco que debe revisar, y es aplicable a partir del 14 de enero de 2027. El fabricante debe asegurarse de que se realice una evaluación de riesgos del producto antes de introducirlo en el mercado o ponerlo en servicio.

Los sistemas de mando deben diseñarse y fabricarse de manera que se evite cualquier situación peligrosa. Un producto conforme con normas armonizadas cuyas referencias se hayan publicado en el Diario Oficial de la Unión Europea se presume conforme con los requisitos esenciales de salud y seguridad que esas normas regulan.

La evaluación por tercero del software solo alcanza a los sistemas con comportamiento total o parcialmente autoevolutivo que usen aprendizaje automático para garantizar funciones de seguridad. Si ese componente ya fue evaluado por un tercero al introducirse por separado, la máquina no tiene que recertificarse solo por incorporarlo.

La máquina debe además recoger pruebas de toda intervención o modificación del software y de su configuración. Según Unión Europea (EUR-Lex), el Reglamento (UE) 2023/1230 presume conformes con los puntos 1.1.9 y 1.2.1 del anexo III a las máquinas y productos relacionados certificados o con declaración de conformidad con arreglo a un sistema de ciberseguridad adoptado conforme al Reglamento (UE) 2019/881 y con referencias publicadas en el Diario Oficial de la Unión Europea, en la medida en que esos requisitos sean objeto del certificado o de la declaración.

Machinery checks before putting an AI system into service on a line

What to check

Rule it rests on

How to verify it

Product risk assessment

The manufacturer must ensure that a product risk assessment is carried out for the product it wishes to place on the market or put into service.

Have the product risk assessment available before it is placed on the market or put into service.

Safety and reliability of control systems

Control systems shall be designed and built in such a way as to prevent any hazardous situation (Annex III, 1.2.1).

If a cybersecurity certificate or declaration exists with references published in the Official Journal, conformity with points 1.1.9 and 1.2.1 is presumed to the extent it covers them.

Self-evolving software with safety functions

Third-party assessment of software only applies to systems with fully or partially self-evolving behaviour that use machine learning and guarantee safety functions.

If the safety component was already assessed by a third party when introduced separately, the machine is not recertified merely for incorporating it.

Conformity with harmonised standards

A product that conforms to harmonised standards, with references published in the EU Official Journal, is presumed to conform to the essential requirements that those standards regulate.

Check that the standards applied have references published in the Official Journal and cover the Annex III requirements to be demonstrated.

Record of software modifications

The machine shall record evidence of any legitimate or illegitimate intervention in the software, or of any modification of the installed software or its configuration.

Check that the record covers interventions and modifications of the software and its configuration.

How does the AI Act affect the classification of an AI system I use in my plant, especially if it affects workers?

Regulation (EU) 2024/1689 requires the provider who considers that an Annex III system is not high-risk to document its assessment before placing it on the market or putting it into service. An Annex III system is always considered high-risk if it performs profiling of natural persons, so it is worth checking this before any other decision.

If a provider considers that an Annex III system is not high-risk, it must document that assessment before placing it on the market or putting it into service, and it must be subject to the registration obligation of Article 49(2). In addition, it must provide that documentation to the competent national authorities when they request it.

The Regulation considers that systems used to manage workers must be classified as high-risk, for example when they allocate tasks based on individual behaviour or personal traits, or when they monitor and evaluate persons in an employment relationship. After consulting the European Artificial Intelligence Board, the Commission was to provide, no later than 2 February 2026, guidelines on the practical application of this article together with a comprehensive list of practical examples of high-risk and non-high-risk AI use cases. In practice, review case by case whether the system is used to allocate tasks based on individual behaviour or personal traits, or to monitor or evaluate persons in contractual employment relationships, uses that recital 57 of the AI Act indicates must be classified as high-risk.

The documentation of the assessment falls to the provider who considers that an Annex III system is not high-risk, and this provider must hand it over to the competent national authorities that request it.

Sources