711 239 085 info@gilsys.com
es ca en
Abstract background

Over 20 years developing software, now serving manufacturing

Artificial intelligence for manufacturing companies

Today's machines generate data that software can access. We bring our development experience to connect them, unify their data, and build AI agents, hand in hand with each company's technical department.

Custom software and AI for industrial machines

We integrate machine data with the same approach to custom programming and AI that we use in projects in other sectors.

Traditionally, a plant's machines have been programmed and maintained by industrial and automation engineers, with the sector's own tools. The most modern machines connect to the network and give access to their data through APIs and industrial protocols. This is where software development comes to the fore. It is about integrating systems and unifying data to build custom applications that can even include artificial intelligence.

Combining our more than 20 years of experience in custom software development with each company's technical department (which knows its machines and processes) leads to high-value monitoring, alerting, and automated reporting solutions.

Phase 1 · Organizing the information

Connecting the machines and unifying data

Each machine stores its data with its own software and in its own format.

Finding out which machine has had the most stoppages this week, or how many parts it has produced, means checking several sources or waiting for a report to be prepared.

Connecting to each machine through the route it offers

Some machines offer an API; others leave the information in a file, a folder, or a database; others expose it through the plant's SCADA or through industrial protocols such as OPC UA or MQTT. How to access the information is studied together with the factory's IT manager.

How to deal with older machines

Connecting an older machine can be complex or even unfeasible. The integration options must be studied with the person in charge. The recommendation is to prioritize machines with protocols that external software can easily access.

A common format for all machines

Once the data is accessed, each manufacturer represents and stores it differently. The data is collected and converted to a common format in a tool custom developed for the plant, with its own API. The tool offers a common access point for the other components. It is the same concept of systems integration carried out between business applications, applied to machines.

A single, standardized database

The data is stored in a standard database such as PostgreSQL, with extensions for machine readings (time series, such as TimescaleDB), for data with a variable structure (JSON), and for search by meaning in manuals and procedures (such as pgvector).

It also allows the plant's own reports to be connected, with monitoring tools such as Grafana or business intelligence tools such as Power BI or Metabase.

Once the machines are connected and their data is in a single platform, it is possible to develop custom platforms that generate real-time reports and alerts, or to create AI agents that can be queried in natural language.

Phase 2 · Artificial intelligence on the data

Queries, alerts, and real-time plant supervision

Artificial intelligence and automations are added on top of the unified data, tailored to each industry.

An AI agent that answers questions in natural language

We can develop an AI agent with MCP (Model Context Protocol), the open standard that lets a language model query data and tools. The agent does not keep a copy of the data and requests from the API the information it needs for each answer. Questions such as "how is each shift performing?" or requests such as "create a chart of monthly production" can be made. We can even cross-reference machine data with shift management tools to find out who was operating the machine when a failure occurred. Each solution can be offered in its own tools or even integrated with applications such as Telegram.

On-demand and scheduled reports

Reports can be generated on demand or automatically, for example with n8n, to send a shift summary to the production manager every morning.

Automatic alerts

Automated workflows, for example with n8n, handle the events. If a machine has been stopped for longer than the set time, a Telegram or WhatsApp message is automatically sent to the maintenance manager. If a temperature goes out of a set range, a light can be switched on and an email sent to management.

Agents that supervise and decide

We deploy AI agents capable of auditing machine data at defined intervals, generating early alerts, and executing automatic actions. In each project we assess how often the information changes and how complex it is in order to apply the most efficient tool, whether rules on time series for direct metrics or AI when complex patterns need to be interpreted.

Real-time dashboards and indicators such as OEE

Dashboards connected to the same data, for example with Grafana, show the status of each machine in real time and can be viewed on a screen in the plant, a computer, or a mobile phone. With the unified data, indicators such as OEE (overall equipment effectiveness) can also be calculated and compared by shift, line, or machine, provided the machines record stoppages, speed, and defective parts.

Permissions per user

In multi-user systems, it is important that the accessible information can be limited according to each user and their profile. Staff in each area must be able to have different levels of access to information. Data that identifies a person, such as who was operating a machine, is restricted to those who need to see it.

Example of a plant workflow with integrated artificial intelligence

How a machine stoppage is handled with unified data

A plant with machines from several manufacturers connected to the same tool. One of them stops mid-morning.

Logging the incident

The machine logs the alert and the tool records it as a "stoppage".

Alert to the phone

After ten minutes, an n8n workflow is triggered that sends a Telegram message to the maintenance manager.

Real-time status

On the plant's Grafana monitoring dashboard, the machine shows a red indicator.

Query in natural language

In the afternoon, the plant manager asks which machines have stopped today, for how long, and why. Using AI, the answer is given in natural language from the data of the different machines integrated into the system.

Data reserved for the plant

The sales team, which can use the same platform for other automations, has no access to this information.

An alert, from the machine to the peopleA plant with three machines from different manufacturers, connected through an API, the SCADA, and OPC UA. Their data is unified in a tool with its own API, which decides who can see what. Machine B stops and the SCADA records it; the tool picks it up as a “stoppage”. After ten minutes, an n8n workflow is triggered that alerts the maintenance manager on Telegram. On the Grafana monitoring dashboard, the machine shows a red indicator. The plant manager asks in the chat which machines have stopped today, and the AI, through MCP, queries the API with the manager’s permissions and answers from the machine data. Sales can use the same platform for other automations but has no access to this information.Which machines stopped today?Machine AAPIMachine BSCADAMachine COPC UAUnified dataTool with an APIPermissionsper userAIMCPChatPlant managern8nand agentsTelegramMaintenanceGrafanaPlant dashboardSalesOther automations123451–5The numbers are the steps of the example in the text.1Machine B stops and the SCADA records it;the tool picks it up as a “stoppage”.2After ten minutes, an n8n workflow alertsmaintenance on Telegram.3On the Grafana monitoring dashboard, themachine shows a red indicator.4The plant manager asks and the AI answersfrom the machine data, with the manager’spermissions.5Sales can use the same platform for otherautomations but has no access to thisinformation.
An alert, from the machine to the peopleA plant with three machines from different manufacturers, connected through an API, the SCADA, and OPC UA. Their data is unified in a tool with its own API, which decides who can see what. Machine B stops and the SCADA records it; the tool picks it up as a “stoppage”. After ten minutes, an n8n workflow is triggered that alerts the maintenance manager on Telegram. On the Grafana monitoring dashboard, the machine shows a red indicator. The plant manager asks in the chat which machines have stopped today, and the AI, through MCP, queries the API with the manager’s permissions and answers from the machine data. Sales can use the same platform for other automations but has no access to this information.Which machinesstopped today?Machine AAPIMachine BSCADAMachine COPC UAUnified dataTool with an APIPermissionsper userAIMCPChatPlant managern8nand agentsTelegramMaintenanceGrafanaPlant dashboardSalesOther automations123451–5The numbers are the steps of the example in the text.1Machine B stops and the SCADA records it;the tool picks it up as a “stoppage”.2After ten minutes, an n8n workflow alertsmaintenance on Telegram.3On the Grafana monitoring dashboard, themachine shows a red indicator.4The plant manager asks and the AI answersfrom the machine data, with the manager’spermissions.5Sales can use the same platform for otherautomations but has no access to thisinformation.
An alert from start to finish. The machine stops, n8n sends an alert, Grafana shows a red indicator, the AI answers according to the permissions, and sales has no access.

We work hand in hand with the technical team

Who contributes what in a plant AI project

The plant

The IT team, the production manager, and the manufacturer or integrator of each machine know their equipment, their network, and their processes. They decide what is connected, how, and where the tool is installed.

Gilsys

We develop custom software, with systems integration, APIs, databases, and artificial intelligence. We create a platform that unifies the data and exposes it through an API. On top of it, whatever each plant needs is custom built, such as real-time dashboards, alerts, automated reports, or an AI agent.

With the factory's IT team

Connecting the machines with AI depends largely on APIs, databases, and software. That is why the project is approached in direct collaboration with the factory's IT team.

As software developers, we do not sell a closed package

What we have ready and what depends on each industry

Platforms and automations are custom programmed for each industry.

Developed and tested in our environment

A central tool with an API, AI with MCP, automations with n8n, agents, dashboards, and permissions. These components are developed and tested in our environment, with test data. In each industry, how to connect each machine, which components to integrate into the final tool, and for what purpose will need to be studied.

Each factory has its own machines and data, so each project is different

Our initial work, together with the company's technology area, will be to investigate how to obtain the data from each machine and how and in what order to integrate it into the platform being developed.

Phase 3 · Preventive alerts

Predictive maintenance with each machine's data

Anomaly detection algorithms are open source. Integrating them into a tool will depend on each machine's data history.

Anomaly detection

With the history of temperatures, vibrations, consumption, and stoppages, the algorithms learn how each machine works and send an alert when it starts to behave differently from usual, before it stops.

Failure prediction

Predicting a specific failure requires a history of well-documented past failures. It is necessary to start by detecting anomalies, and prediction is added once that history exists.

Gradual integration

It is advisable to run a test on a specific line or machine, measuring correct detections and false alarms. Those figures show whether it is worth extending it to the rest of the plant.

Results in the same platform

Anomalies are recorded in the same platform as the rest of the machine data, trigger alert workflows, appear on dashboards, and can be queried in natural language by interacting with AI.

Phase 4 · Under study

Computer vision for quality control

At Gilsys we are opening a line of research to make it possible to integrate computer vision into custom-developed tools.

A test with photos of real parts

An open-source detection model recommends whether each part is correct or defective and is compared with human inspection, without intervening in the line. For infrequent defects, it can be trained with correct parts only.

What the plant provides

The camera, lighting, and line signal are provided by the plant or its integrator. Whether the images can be processed at the plant itself is under study.

Results in the platform and in the MES

Each result (correct or defective part, time, and line) goes into the data platform, appears on the dashboards, generates alerts, and can be queried in natural language. If the plant has an MES, the information is also sent to it through OPC UA, MQTT, or a REST API.

Tools for digitalizing a manufacturing company

Beyond machine data, a manufacturing plant shares digitalization needs with any other company.

Document management

Drawings, technical data sheets, and quality certificates can be organized in a document management system on the company's own servers, with permissions by role and search by content.

Product catalogs with Pimcore

A PIM such as Pimcore classifies large product catalogs together with their digital assets (images, technical data sheets, or manuals). It also syncs them with online stores, marketplaces, the ERP, and other external platforms.

High-performance corporate websites

A high-performance corporate website can display products with the information from the PIM without duplicating it, while improving the company's ranking in search engines (SEO) and in AI assistants (GEO).

AI automation

Orders, delivery notes, and supplier invoices can be interpreted through AI automations and recorded in the ERP, with review by a person when the AI is not certain.

Private AI

When drawings, processes, or documents must not leave the company, or work must be done offline, a private AI can run on the company's own servers with open-source models.

Custom software with AI

Custom software with AI can search a technical catalog by description rather than by exact reference, or offer the sales team a chatbot with that catalog.

Custom calculation tools

The company's technical calculations can be integrated into a custom web application that applies advanced mathematical algorithms and generates comparative performance charts.

Data governance

Before an AI project, it is advisable to carry out a data quality audit of the ERP that reviews duplicates, invalid values, and structure.

AI consulting

AI consulting gives management an assessment of where AI can add value and a phased roadmap.

Although every company is different, the workflow is similar

How we organize an AI project in a plant

For plants in Catalonia, the machine review is carried out on site. For other areas, scope and availability are checked.

Machine review

What data each machine provides and how it should be accessed, in consultation with its supplier.

Choosing the first case

One or two machines and a specific need, such as an alert, a dashboard, or natural language queries.

Connecting and unifying data

We develop the connection with the machines and the tool that unifies their data, with its API.

AI, automations, and dashboards

We add the AI layer, automated workflows, agents, and dashboards, with each user's permissions.

Expansion and adjustments

We add machines and automations and adjust whatever is needed as it is used.

Frequently asked questions about AI in manufacturing

Do the machines need to be replaced?

The connection uses what each machine already allows, such as an API, the SCADA, OPC UA, MQTT, or the files it generates. Only an older machine with no connection at all may need a gateway or a sensor.

Does the plant data leave the company?

Not necessarily. The tool and the AI can run on private servers, with open-source models, as in a private AI setup. Even so, since this is technical data, working with commercial AI models is not usually a problem.

Is it safe to connect the plant network?

As proposed, the connection with the machines is read-only, which means their data is read and no commands are sent to them. The tool is installed wherever the plant network manager decides, and it is coordinated with the manufacturer or integrator when access goes through them.

Does it connect with the plant's ERP?

It is possible to study integrating machine data with the customer orders and production orders in the ERP, through its API or its database, as in any systems integration. This makes it possible to report, for example, which order was being produced when a machine stopped.

How much does a plant AI project cost?

It depends mainly on how many machines need to be connected and how each one connects. A free, no-obligation estimate can be provided.
Abstract background

All the plant's data in a single system

It is enough to indicate which machines are in the plant and what needs to be known or automated. The data they provide is reviewed and a first case is proposed.