What does the Azure platform offer and which advanced Azure AI services are ready for production projects?
Humans have always tried to understand the world around us, using tools that are extensions of our brains and expand our capacity to think and understand. In the past, these were more primitive tools, but as computing and science have evolved, we are also using more and more advanced tools and services that can help us solve complex problems with targeted domain knowledge, using approaches from artificial intelligence and machine learning.
Artificial Intelligence in the Microsoft Azure Cloud
The Azure cloud offers a set of advanced Azure AI services that can help us solve these kinds of problems and are technologically ready for production projects. On the one hand, we have the SaaS solution Azure Machine Learning, which allows development engineers and data scientists to perform the entire machine learning process using advanced graphical interfaces. This enables, for example, data labelling, data preparation (transformation, cleansing ... ) and the preparation of a dataset that serves as input to machine learning methods. On the other hand, we have a set of solutions called Azure Cognitive Services, which enable a variety of intelligent processing through an application programming interface (API):
- machine translation of text,
- speech to text,
- image and video recognition (Custom Vision, Face API).
One of the proofs that Microsoft's AI solutions are world-leading and especially ready for non-data science end-users is this year's Gartner report, which ranks Microsoft as one of the leaders in the field. Azure Cloud AI Developer Services is a leader in completeness of vision.1
In the following, we will present one of the solutions that is part of the intelligent video processing.
Data is the new crude oil, and tools like these allow us to really focus on the data and its preservation, ensuring that our prediction and classification models are accurate and that they allow our clients to improve their knowledge of a given problem.
In the stroka.si Group, the Research and Development Department has successfully completed several major projects in the field of machine learning using Azure services, including:
- IQ TPM 4.0 – predictive maintenance in production, using Azure AI, provided the flow and collection and tagging of data in a way to perform the prediction of the possibility of machine downtime (CNC machines and forging hammers) in production.
- biziSMART – sales forecasting, which uses Azure Machine Learning for companies in the Bizi.si database of business entities to indicate the possibility of selling services to one of the subscribers (the solution was developed in cooperation with the partner TS Media).
Structured data storage and strokaDT platform solutions
Proper structuring and efficient and reliable data storage are crucial in this type of project, which is why AI and machine learning technologies are in practice just an extension of existing systems (ERP, CRM, MES, IoT systems, etc.) and are often referred to as complementary technologies. It is extremely difficult for them to exist on their own, as they require a mass of data to function, which can then be used to perform the machine learning process itself, leading to predictive models. For this purpose, the stroka.si Group has developed its own system, which we call the strokaDT platform, and which serves as the core of all projects for the digital transformation of companies.
Figure 1: podkaDT - a platform for building digital transformation solutions
The solution establishes procedures for the flow of data throughout the platform and acts as an aggregator of the various data sources (ERP, CRM, MES, etc.) in the company, thus ensuring:
- appropriate structuring and long-term storage of data (Microsoft SQL, Azure BLOB, CosmosDb, Azure hosted InfluxDb), ready for further operations and preparation of machine learning methods;
- integration of subsystems using strokaAPI, which enables the integration of different data sources in digital transformation projects;
- data visualisation using tools such as Microsoft PowerBI and Azure Managed Grafana, and dedicated visualisations developed using Microsoft Blazor technology;
- Hybrid data storage, which in practice means that the entire infrastructure can be hosted in the Azure cloud, or specific sections can be moved to the company's internal infrastructure.
Azure Percept and machine vision solutions
As we closely follow the development of Microsoft Azure cloud services in the stroka.si Group, we were probably the first in Slovenia to test the Azure Percept solution. It is a combination of hardware (Azure Percept DK) and a cloud-hosted intelligent system Azure Percept Studio, which performs machine learning in the field of machine vision with the help of the customvision.ai solution.
It is therefore a combination of cloud and edge computing, as the predictive model for object detection in the video is installed and runs directly on the Azure Percept appliance, a collaboration between Microsoft and Asus. The purpose of the testing was to see if Azure Percept could be used on any of the Group's client development projects. We tested a general object detection model learned from the most common objects found in the office.
The results and recognition performance are quite good, but we were most impressed with the real-time vehicle recognition model as shown in the screen capture below, which is the view from the stroka.si Group office (through a tinted window, which in some ways makes vehicle recognition even more difficult).
Figure 2 - Real-time vehicle recognition in vision using Azure Percept
The hardware part of the Azure Percept solution has been integrated into the in-house solution of the strokaDT platforms by sending telemetry and detection data from the device to the Azure cloud. The Azure IoT Hub serves as the entry point, as the device has been integrated into the internal system of the strokaDT platforms. During the test, our developers also continued to learn an internal model that can identify randomly selected objects from the office.