Artificial intelligence (AI) may be a helpful tool in various sectors, especially where large bulk of data is processed. One of the major benefits of AI is the automation of routine tasks contributing to acceleration of processes and reduction of labor costs. Moreover, AI may significantly increase the accuracy and safety of decision making processes in various spheres of activities. Industry sector is not an exception. Let's see how manufacturing companies use AI.

Computer Vision is used in conveyors to define the products' elements, their abnormalities and geometry. Thus, it levels out the human factor effect and reduces the amount of rejected products.

Another critical application of AI is a customer experience management. AI helps generating customized proposals, monitoring the trends and seasonal changes, as well as improving the logistics and purchases.

It should be emphasized that each company is keen on developing its own unique solution, since a better result requires additional training of models on the client's data.

AI integration is a complicated process requiring certain competences and resources. First, to ensure successful AI integration a company shall have an in-house expert review or know which specialists it needs to involve. Additionally, access shall be provided to a large bulk of data used for the models training. This means not only the need for data availability, but also their labeling and making them ready for training.

One of the challenges is searching best tasks that can be solved using artificial intelligence. For each of the areas, such as Computer Vision, Data Science or language models, it is necessary to formulate the specific tasks, understand the subject and have statistical data to tackle the problems in the most efficient way.

Another issue is the uncertainty of the neural networks output especially when dealing with text models. The statement provided by the model cannot always be predicted which may be a problem in some cases.

AI integration into existing systems may also be problematic and require additional efforts and resources. AI integration may constitute a sufficient part of the project budget.

A point to keep in mind is that none of the AI-based solutions can boast absolute accuracy, therefore, the role of the human and operator shall remain, particularly in highly critical areas.

However, despite the challenges, AI integration has many benefits and a potential for processes optimization, products quality improvement, automation of routine tasks and higher operation efficiency in various business activities.

Language models, such as GPT, may be used for automatic generation of reports, facilitation of search and provision of expert guidance, thus, contributing to a much better cooperation with the clients and data processing. Due to the use of Artificial Intelligence in search of appropriate solutions, integration and data usage, the companies may achieve tangible results in their business activity.

Let's have an insight into specific cases implemented by Globus IT in the industry sector.

GPT-models integration into the company's loop

Client – a large producing company.

Task. Secure GPT-models integration into the company's loop with additional models training on specific topics, documents and databases and building of individual business-processes.

Solution. InsightStream smart search.

Access to the information through a smart search accomplished as a corporate knowledge database chat.

— Relevant response — summation of facts from the knowledge database with links to the primary sources.

— Processing of multimodal data and integration with various data storage systems.

— Structuring and uniting of scattered data.

Areas where InsightStream was helpful:

1. Self-service for employees: resolving issues on processes and regulations.
2. Removal of excessive document flow and filling the gaps in the knowledge database.
3. Faster work of business analysts in the product teams.

Areas where InsightStream may help:

— Long go-to-market and settlement of incidents resulting from the lack of data.

— Poor quality of content in the knowledge database: duplicated data, missing data, non-relevancy, and data presentation.

— Poor study skills of employees and long time to peak efficiency

Smart assistant VirtualGuru

Automation of expert guidance and in-house/external support

— Analysis of the incoming intents as a structural representation (requests).

— Automation of the answers to questions in the human-in-the-loop-scenario of integration.

— A mix of document search, Web search and LLM-model response to provide enhanced covering of topics.

Areas where VirtualGuru proved to be good:

1. Improved efficiency of a legal service agency due to the use of «legal adviser» chat-bot.
2. Automated processing of in-house service requests: HR, lawyers, finance…
3. Routing of complex intent messages in the support service.

VirtualGuru may tackle the following:

— High costs of inner and outer requests processing.

— Unsatisfactory metrics: long response, «bounce» of request, low NPS.

Result

1. Web-service. Access via browser. Company's corporate portal window with the assistant interface.
2. Chat-bot. Chat-bot-assistant for the Company's employees in Telegram or corporate messenger.
3. Integration with API. Access for any IT-product. Coordinated API-interface for integration by the developers.

Indicators:

— Information search rate increased up to 10 times.

— The goods delivery speed raised by 30%.

— Support assistants became 5–7 times more efficient.

— Up to 95% of compound letters were correctly routed.

Occupational safety improvement at a large oil refinery

Task. Monitoring of safety requirements compliance at the refinery's facilities: ACS, regulations for wearing overalls and etc.

Solution. Real-time commercial analytics based on AI, ML, CV and neural networks.

The system recognizes and analyzes the images from cameras installed at the facility site, compares the images with the database patterns (for ACS – database of photos of authorized employees; for overalls control – availability of logos and tags), interprets the results and informs the responsible persons in case of violated regulations.

Technologies: Python, Computer Vision, Machine Learning, Deep Learning, Object Detection, Image Recognition, Multiple Object Tracking

Tracing of rejected products

Client – electrical equipment manufacturer

Task. Visual tracing of rejected products and correctness of technical information on them (stickers, labels, tags).

Solution. System of objects video analysis and interpretation.

Video processing using OpenCV, objects tracing, stickers’ area highlighting. Use of natural language processing (NLP), convolutional neural networks (CNN) to select, process and compare the text with the pattern. Use of deep learning (DL) methods to check the correctness of stickers’ location.

Technologies applied: OpenCV, ML/DL, Python, Keras, TensorFlow.

Result. Final savings from the technology deployment amounted 4%.

Products analysis and template matching on conveyor using a set of templates

Client — A company integrating the products automated control on conveyor.

Task. Analysis and comparison of products on conveyor systems using a set of templates.

Solution. A system of objects video analysis and interpretation was developed.

The video is processed using OpenCV, objects tracking, objects areas selection. To identify the problems and deviations a wide-range classification ML-models (SVM, K-NN, Decision Tree, Random Forest, AdaBoost, LDA, K-Means) for a quick search of objects and objects types and a wide range of DL-models for the object checking and template matching are used. The system is integrated with the conveyor control systems to provide an emergency stop function.

Technologies applied: OpenCV, ML/DL, Python, Keras, TensorFlow, Java, Angular, Oracle DB, ELK stack, Docker.

Result

Automated production process, reduced the amount of rejected products by 5 — 10%, minimal products defects when retrofitting the conveyor for a new type of product. Great reduction of the products final costs due to deployment of automated processes and lower number of defective products.