For Development Teams: Anyone looking to develop production-ready AI for machines needs a powerful MLOps runtime environment to train, deploy, and monitor AI models – essentially the engine of an AI application. Standard frameworks aren’t built for industrial machine environments, which is why aiXbrain has developed Dataray as an optimized “best-of” package exactly for that purpose. So why spend time and money building from scratch when you can get a tailor-made, cost-effective solution?

For Service & After-Sales Teams: Providing premium service and innovative add-ons is nearly impossible today without AI. With Dataray AI Apps, you get access to advanced wear prediction, anomaly detection, and root cause analysis – fully integrated into your data and monitoring platform. This enables you to resolve issues at unprecedented speed, optimize spare parts management, and offer customers digital solutions to enhance machine performance.

A Dataray license corresponds to a Dataray instance, which allows you to train, deploy, and monitor one active AI model. In general, a single instance can serve up to 50 machines. However, the actual number of machines a model can support depends on the statistical variance of the machine data and the required accuracy of AI results. In most cases, multiple identical machines can be operated using the same active AI model within a single instance.

The onboarding takes place in four steps. First, during the AI specification phase, the required AI input and output are defined, quality requirements are set, and the first suitable AI models are selected. This is followed by the configuration phase, where a Dataray instance is set up and the AI model is trained. After a successful evaluation, the guided rollout begins, including regular feedback sessions and an analysis of usage statistics. In live operation the AI application is continuously monitored and updated, with aiXbrain experts available for support if needed.
If Dataray is not yet integrated into the existing data platform, an additional integration step is required before onboarding.

Generally, none. While a basic understanding of what AI can do and which data it requires can be helpful, it is not a prerequisite. Dataray comes with pre-configured AI models designed for industrial applications and is fully supported by our expert team during setup and operation. Advanced users can independently monitor and adjust AI models and collaborate with us to further develop models for specific use cases.

All customer and machine data remain the exclusive property of the customer. If metadata or usage data is stored in Dataray, these also remain sole property of the customer. AI models trained on your data can be used without restrictions, both internally and in applications for your end customers. Our customers receive an exclusive, perpetual, and unrestricted usage right for the executable AI models trained with their data. Upon request or at the end of our collaboration, the executable code of these models is fully deleted.

Dataray directly accesses machine and process data from your connected data platform. To accelerate AI processing, data can be pre-processed and stored in an optimized format if needed.
Additionally, Dataray only stores configuration settings, usage data, and metadata that cannot be directly stored in your platform.
For cloud deployments, all data is stored exclusively on EU servers, fully compliant with applicable data protection and security standards. This ensures maximum performance, transparency, and data security.

Dataray uses an optimized set of proven AI models for industrial applications, such as analyzing low-frequency control data or high-frequency sensor data. During training, multiple models compete, and the best-performing model is selected as the active AI model. For time-series data, Dataray primarily relies on Deep Learning models, such as neural networks. For image data, it leverages pre-trained models, and for text-based tasks, it uses LLMs and Generative AI technology.
As AI experts, we continuously evaluate new models and add them to Dataray’s AI Model Repository when they provide additional value.

Modern pre-trained models and Generative AI require very little training data. For other Deep Learning models, the required data volume depends on the desired AI accuracy, the statistical variance, and the informational value of the training data. A small number of relevant events, such as rare machine failures, can result in large datasets that are unsuitable for training. As a general guideline, we start with several hundred to a few thousand data points, with at least one-third representing relevant events.

No. Whether labeling is required depends on the task and the desired AI functionality.

Supervised learning requires labeled data to provide the AI with clear patterns and target values. This is typically applied to classification or fault detection models.

Unsupervised learning, on the other hand, works without labeled data and identifies patterns on its own, such as for anomaly detection or clustering.

In some cases, a hybrid approach can be beneficial, where AI is initially trained with a small amount of labeled data and then continues learning independently.