How can ROS 2 and industrial AI be used in your business?

What can ROS 2 bring to our business?

ROS 2 helps consolidate information and software tasks from different devices into a common layout. For example, the camera recognizes a part, the system matches this information with the relevant work order and the action to be taken by the robot is determined. In order for the devices to coordinate the same work, the connections between them are developed according to the project.

In a hypothetical part-sorting scenario, a camera could publish each part's identity and analysis result through a topic. The control software could request the station's current state through a short service call. A longer task, such as moving the part to another location, could use an action interface to track progress, results and cancellation. The interface choice is assessed alongside the actual application's control and safety boundaries.

A well-planned system can make it easier to monitor the status of different stations and add a new device in the future. The maintenance team can see more clearly which part is working and where the problem occurs. For this purpose, data recording, error reporting and maintenance screens should also be designed. ROS 2 is not required on every machine; If a simpler solution meets the need, we will also evaluate it.

• Sharing information and tasks between devices
• Monitoring production flow and problems
• Considering future additions from the outset

In which jobs can it be used in the factory?

On an assembly line, a camera can assess the type or placement of an incoming part. The result is linked to the correct part record so the robot receives the appropriate task. Unrecognised or uncertain parts are referred for operator review. This illustrates an application in which image analysis is designed together with the next production step.

In another scenario, temperature, vibration or cycling information on machines can be collected. AI supports evaluating unusual patterns; The maintenance team makes a decision by reviewing the records. It is also possible to combine production and quality information from multiple stations. These are examples of project scope; It is not a product that is ready for every line and works the same way. The goal is clarified through experiments with device accessibility and real data.

• Matching part recognition and robot task
• Transfer of machine data to maintenance evaluation
• Production and quality tracking between stations

Can our existing machine, PLC and robots be used?

At the beginning of most projects we examine what can be done while preserving existing equipment. Industrial computers, suitable small computers, cameras, sensors, custom electronic circuit boards and server software can undertake different tasks of the system. Not every device needs ROS 2 running in it; Exchanging information over the appropriate connection may be sufficient.

The existing PLC or manufacturer-provided connections and software access for the robot are checked. Compatibility cannot be promised without seeing the device model, program and available data. PLC can maintain current control of the machine; The new system can monitor, evaluate or share high-level tasks. Safety functions remain within the appropriate control and safety architecture determined by the risk assessment. It also defines how the machine will behave when the connection is lost or the new system becomes unresponsive.

• Reviewing existing equipment and manufacturer-provided interfaces
• Development of custom electronics or connection software if necessary
• Determination of control, maintenance and safety responsibilities

What is the difference between centralized management and distributed system?

In the centralized system, information comes to the main computer and a significant part of the work is carried out there. For example, the results of several stations can be evaluated on the same computer and displayed on a common screen. It may be useful to centralize maintenance and updates; however, the impact of this computer or connection being disabled should be planned for.

In a distributed setup, some work is done on devices next to the station. For example, the image is evaluated on the local computer, and the result and production information is sent to the center. Thus, it may not be necessary to always collect all the images in the center. The two approaches can also be used together in the same project. The decision is determined by the number of stations, the amount of data, network conditions and the work to be continued during the outage. What data authorized users will access and how remote maintenance will be carried out are also decided.

• Evaluation and follow-up in a single center
• Local operations next to the station
• Task sharing in accordance with outage and maintenance conditions

Does artificial intelligence require CPU or GPU?

CPU is the general processor of the computer. GPU is a graphics processor that can be used to run many similar calculations together. Some AI models meet the need on CPU; For more intensive work, GPU can be considered. Purchasing GPU does not in itself mean that you get a better or faster system.

Training a model on examples and using it in production are different tasks. A model trained on a powerful computer may run on a smaller device where suitable. We choose hardware by testing actual production data on candidate devices. Result quality, required response time, heating during extended operation, energy use and maintenance are assessed together. Identifying hardware that meets the need helps avoid unnecessary computer investment.

• Choosing a processor according to the needs of the job
• Separate planning of training computer and field device
• Checking quality and execution time with real data

How to develop artificial intelligence from business data?

The first step is to collect examples that are representative of the task to be taught. Good and defective parts, different products or normal and unusual machine records are evaluated by experts. The model is trained with these examples; It is then tested with recordings that were not used in training. Which errors are unacceptable are determined by the quality and production team of the enterprise.

Closed-loop AI development brings new field examples back into the development process. Authorised people review them, approve corrections and prepare a new model. The new version is checked against earlier samples and released after limited testing and human approval. It is not expected to change uncontrollably by treating its own decisions as correct. Model versions and deployment locations are recorded, with rollback when needed. Product or environmental changes trigger reassessment.

• Samples belonging to the business and examined by experts
• Separate testing and human approval for new model
• Registration of the version used and reverting to the previous model

How can such a project be introduced while production continues?

First, the existing line, devices, software and production flow are examined. To start, choose a clear-cut task: evaluating a group of parts at a single station or collecting specific machine information. The expected results and the conditions under which acceptance will be given are written together. The current situation is measured; trial results are compared to this.

Initial experiments can use recorded data or a test setup separate from production. Where needed, the workflow is simulated on a computer; this does not replace actual line testing. A monitoring phase may then let the new system produce results without sending machine commands, so its recommendations can be compared with the existing process.

In limited pilot, disconnection, bad data, restart and operator intervention are also attempted. Commissioning time is planned with the production team; The path to return to the old working order is preserved when necessary. After acceptance, maintenance steps and user responsibilities are explained. Transition to other lines is made by evaluating the first pilot's records and each line's own conditions.

• Current situation analysis and first small-scale trial
• First monitoring, then controlled piloting
• Acceptance, maintenance and rollback plan

How are our data and project files protected and delivered?

VOMRO signs a confidentiality agreement (NDA) before sharing technical details upon your request. Who will access the data, where it will be stored and whether external services will be used are determined from the beginning. If the resources and selected components are suitable, data collection, training and operation of the model can be planned in-house.

Source codes of the project, developed model files, electronic schematic / PCB and mechanical CAD resources and technical documents prepared within the scope are delivered at the end of the project. Installation, connections, versions used, test results and update steps are also included in the delivery list. Licenses for third-party components are disclosed separately. Our PCB production and electronic component assembly support is within the scope of customer projects we develop. You can start the first meeting by explaining your current equipment and the problem you want to solve.

• NDA before sharing technical information
• Determining data access and storage conditions
• Source code, model, PCB/CAD and technical documentation delivery