Many companies have committed to climate neutrality and want to make their contribution to climate protection and greater sustainability. Concrete measures on how to reduce the emissions caused by their own company (whether directly or indirectly) must be implemented. Companies' strategies for greater sustainability vary greatly. Each company must find an individual solution, be it monitoring and optimizing production to reduce waste, saving energy in the operation of process plants or the intelligent monitoring of production processes. Artificial intelligence (AI) can be a tool that can be used to optimize production processes and save resources such as energy, materials or additives. All potential savings that are tapped into contribute to the carbon footprint and sustainability.

AI can influence sustainability goals in two ways: Firstly, AI can be used as a technology to make production more sustainable; secondly, it is possible to observe the extent to which an AI solution itself is sustainable - i.e. how resource-intensive the development and operation of the AI solution is.

Optimizing the welding process with AI

By optimizing production processes using AI, for example, the quality of the products can be ensured and waste reduced. Less waste means less material is used and less energy is required.

The internationally active Grenzebach Group, for example, relies on Weidmüller's Industrial Analytics software. The project focuses on monitoring the condition and quality of the friction stir welding process. Sensors installed on the machine record the process data during the welding process and, together with the analytics software, enable real-time monitoring of the welding process. Intelligent data analysis enables both precise quality assurance and targeted maintenance of the machines, thereby saving resources.

In friction stir welding, metals are stirred rather than melted at the joint using a rotating tool. The joint is heated and joined in a solid state.

© Grenzebach

With its friction stir welding systems, Grenzebach offers a joining process for joining aluminum. It reproducibly joins workpieces in a pressure- and media-tight manner. Metals are stirred at the joint using a rotating tool and - in contrast to arc and laser welding processes - are not melted. The joint is heated but joined in a solid state. Friction stir welding can be optimally integrated into industrial series production.

Up to now, quality control in this process has traditionally been rule-based and carried out visually and manually by the operator after the welding process on an as-needed or random basis. An AI solution enables more efficient control: the trained machine learning (ML) model compares the recorded process data with a reference data set during the welding process. As soon as there is a deviation outside of the defined parameters, the machine operator is notified of an anomaly in the welding process.
The analytics software makes it possible to evaluate the process parameters and therefore also the quality of the weld seam and the parts produced during the welding process - and not afterwards, as was previously the case.

Focus on saving energy

Many sustainability projects focus on saving energy. This does not only apply to industrial production. Energy can be saved in many other areas of application through the use of AI, for example in water treatment in sewage treatment plants. After the initial mechanical separation, i.e. the removal of solid substances, the water is slowed down in the subsequent process so that further pollutants can settle. The most energy-intensive wastewater treatment process then begins in the aeration tank. This is where the biological purification of the wastewater begins. Bacteria break down harmful compounds in an alternation of aerated and non-aerated tank zones. Air is blown into the tank so that the bacteria can work and breathe. This makes the blower one of the main energy consumers in a sewage treatment plant. ML methods can be used to determine an "intelligent mode of operation" for the blower, i.e. the optimum operating point in terms of load requirement and energy consumption. The ideal setting of the blower can save up to 20 % in energy costs, which leads to a considerable_{reduction in CO2}. Ecological and economic sustainability go hand in hand.

How sustainable is AI?

The positive sustainability effects of using AI are offset by the energy requirements of AI solutions. The training of AI models in particular is very energy-intensive. During model training, numerous model variants are created and optimization loops are run, which places a corresponding strain on the IT infrastructure and results in high energy requirements, among other things. Model execution during operation and the management and updating of models also consume corresponding resources.

For sustainable AI, energy efficiency must be taken into account as an additional goal when training the models. The energy required to train and operate the model must be set in relation to the quality of the model. This allows an AI model to be assigned a_{carbon footprint} and the impact of AI on sustainability goals to be assessed holistically. The sustainability of an AI model is determined in particular during the training phase. The fewer iteration loops run during training and the less data is required for training, the more energy-efficient the training process is. It is therefore important to converge the solution space as quickly as possible in order to only pursue those AI methods or run optimization loops for those models whose optimization also appears worthwhile - i.e. leads to an improvement in model quality. Similar correlations also apply in principle to the operating phase: the less data the model has to process, the more efficient the model is and the less frequently the model has to be executed, the fewer resources are required.

The sustainability of an AI model can be influenced by domain knowledge. So far, the importance of domain knowledge in the training process has primarily been discussed due to its influence on model quality. Domain knowledge can also be used to improve the energy balance of AI models. To this end, Weidmüller's Industrial AutoML software offers the user the option of restricting the solution space for model training, e.g. by creating or selecting domain-specific features for model training. This sets boundary conditions that enable an efficient run through the solution space during the training phase. The aim is to achieve an AI solution with a better energy balance over the entire life cycle.

AI brings many advantages

The author: Tobias Gaukstern is Head of Business Unit Industrial Analytics at Weidmüller.

© Weidmüller

When used correctly, AI is a catalyst for ecological and economic sustainability. AI helps to save energy and reduce_CO2 and enables more efficient use of resources such as water, materials, consumables, machines and much more. At the same time, the use of AI can increase productivity in production.