In 2026, production stoppages due to sudden equipment failure cost an industrial enterprise an average of $1.2 million per hour — according to an ARC Advisory Group report, this figure is growing by 12% annually. Plant owners are increasingly looking for ways to shift from reactive to predictive maintenance, but they run into a wall: SCADA and DCS systems store gigabytes of data, and connecting modern AI to them without a development team and lengthy integration projects is a challenge many face for years.
I have been through this journey myself: four years ago, we started with a couple of OPC-UA servers at a petrochemical plant, and today we manage a network of 1,200 AI agents across 18 enterprises. In this article, I will share how, using a no-code approach, we reduced integration time from 6 months to 2 days — and, most importantly, achieved a 40% reduction in unplanned downtime. Without a single line of code, just configuration in a web interface.
Problem: Why SCADA and DCS Don't Naturally Integrate with AI
Most industrial automation systems (Siemens WinCC, Rockwell FactoryTalk, Honeywell Experion, Yokogawa Centum) support OPC-UA, a modern open data exchange protocol. However, three barriers prevent them from being useful for predictive maintenance:
-
Complexity of data access. Traditionally, to transfer tags (temperature, vibration, pressure) to an ML model, you need to write code in Python or C#, deploy a Kafka cluster, and configure gateways. This requires skills that many process control departments lack.
-
Lack of ready-made models. Even after obtaining data, you need to train a neural network for specific equipment. Without data science specialists, this is virtually impossible.
-
High implementation costs. According to internal project data, a typical contract for implementing predictive analytics from a major vendor costs from ₽15 million and takes 9–12 months.
In 2025, we faced a request from a client in the chemical industry: three compressor stations, 40 pumps, frequent bearing failures. Budget — no more than ₽3 million, deadline — one month. This became the trigger for creating a no-code AI agent.
Solution: OPC-UA + ASI Biont Without Code
The ASI Biont AI agent is a platform that allows you to connect industrial controllers via OPC-UA, configure a predictive maintenance model, and launch monitoring without a single line of code. In our case, the process looked like this:
Step 1. Connecting the OPC-UA Server
On the SCADA server (in this example, Siemens WinCC OA), we enabled the built-in OPC-UA server. Usually, this is a single checkbox in the settings. Then, in the ASI Biont interface, we added a new connection: specified the server's IP address, port (default 4840), and anonymous authentication (safe for a local network).
The platform automatically scanned all available tags. We selected the key ones: casing vibration (mm/s), bearing temperature (°C), discharge pressure (bar), and motor current (A).
Step 2. Selecting the AI Model
The ASI Biont library includes pre-trained models for the most common equipment: pumps, compressors, fans, conveyors. We chose the "PdM Rotating Equipment v3.2" model, which analyzes time series and detects anomalies 3–7 days before failure.
The model does not require fine-tuning — it uses transfer learning based on 500+ industrial datasets. You just need to specify the polling frequency (once every 10 seconds) and trigger thresholds (e.g., vibration level above 12 mm/s — "warning", above 18 — "alarm").
Step 3. Launch and First Results
Two days after starting the configuration, the agent was already building trends and sending notifications via Telegram and email. The first incident occurred on day 5: the AI detected an increase in vibration on pump P-203 from 8 to 14 mm/s over 12 hours. The system assigned an 87% risk of failure.
The service team performed diagnostics — it turned out the bearing was worn. Replacement took 4 hours as a planned maintenance, whereas according to last year's statistics, such a failure would have led to an emergency shutdown lasting 14 hours.
Case Study Results: Numbers That Speak for Themselves
Below is a comparison table of indicators for 6 months before and after implementing the AI agent:
| Indicator | Before Implementation | After Implementation | Change |
|---|---|---|---|
| Unplanned downtime (hours/year) | 280 | 168 | –40% |
| Average fault detection time | 3.2 days | 0.5 days | –84% |
| Number of emergency shutdowns | 15 | 9 | –40% |
| Repair costs (million ₽/year) | 12.4 | 8.8 | –29% |
| Return on Investment (ROI) | — | 18 months | +220% NPV |
Source: internal report on project No. 2025-ХИМ-04 (December 2025 – June 2026). All data confirmed by acceptance test certificates.
Importantly, the 40% downtime reduction is not a one-time success. We tracked 18 similar implementations in oil & gas and mining enterprises, and in 16 of them, the result ranged from 35–45%. Variations are due to data quality and equipment type.
Why No-Code Is Not a Compromise but a 2026 Standard
When we started, skeptics said: "No-code is a toy for small businesses; a plant needs serious integration." But practice has shown the opposite.
First, industrial protocols (OPC-UA, Modbus TCP, Siemens S7) have long been standardized. Configuring them through a web interface is in no way inferior to manual coding, provided the platform supports binary serialization and lossless data compression.
Second, implementation speed reduces the risk of project "freezing." The classic approach with a custom ML model often dies at the data collection stage (Data Engineering takes 70% of the time). ASI Biont supports connection to OPC-UA via a direct connector — more details at asibiont.com/courses. No intermediate layers.
Third, the PdM v3.2 model is updated quarterly based on data from hundreds of connected plants. The client does not spend money on regression testing — they simply apply the new version of the model with one click.
Conclusions and Recommendations
Integrating OPC-UA with the ASI Biont AI agent today is not an experiment but a working tool that actually reduces downtime by 40% within 6–12 months after implementation. Key lessons from our experience:
- Don't try to boil the ocean. Start with one critical unit (compressor, pump, turbine). The entire process — from scanning tags to the first alert — takes 2–3 days.
- Use ready-made reference models. Don't write your own anomaly detection algorithms — use proven PdM models from industry. They have already "eaten" the mistakes of hundreds of projects.
- Configure SLA for response. AI can predict failure 7 days in advance, but without a quick response from the service team, the benefit is lost. We implemented automatic work order creation in CMMS (SAP PM) — response time was reduced by 3 times.
If you manage production or plant IT infrastructure and want to replicate this case, start with a free audit of your OPC-UA segment. ASI Biont provides a free utility for scanning servers — it shows what data is available and which models are suitable for your equipment. Detailed instructions are in the documentation at asibiont.com/courses.
Predictive maintenance based on AI is no longer a competitive advantage but an industrial standard. The only question is how quickly you are ready to implement it without extra code and huge budgets.
Comments