What is industrial data analytics and why do you need it?
Industrial data analytics means regularly collecting the raw signals a plant's machines produce (whether a machine is running or stopped, part counters, temperature, pressure, energy consumption, cycle time and so on) and turning them into information people can interpret. Many factories already have this data, but it is scattered: some of it stays in the machines' programmable logic controllers (PLCs), some appears briefly on operator screens and is gone, and some is written by hand into Excel sheets or on paper at the end of the shift. Answering questions such as "Why was the line slow yesterday?" or "Which machine stopped most often this month?" then takes days, and the answer is often a guess. With data analytics in place, a dashboard answers the same questions in minutes. Downtime reasons, scrap rates and capacity utilization are tracked shift by shift, and problems are caught before they grow. Production managers can then base investment, maintenance and planning decisions on measured data, and discussions revolve around a shared figure rather than personal impressions.
How does machine data reach the dashboard?
Machine data reaches the screen through five connected layers, each with a clear job. The first layer is the data source: PLCs, sensors and machine control units. In the second, this data is read using industrial communication protocols. OPC UA is an open standard that lets machines from different manufacturers speak a common language; Modbus is a simple, widespread protocol used on the shop floor for many years; MQTT is designed to carry sensor data to a central system in lightweight messages. In the third layer, measurements are written to a time-series database as timestamped records, a structure well suited to querying millions of measurements quickly. The fourth layer is the dashboard, where data is summarized by shift, machine and product. The fifth is the alerting system, which notifies the right person when a value crosses a threshold or an unusual pattern appears. New sensors are not always needed; many modern machines already hold the basic information in their PLC. Step by step, the flow works like this:
- Source: PLCs, sensors and machine control units; signals such as running status, counters, temperature and current.
- Collection: reading the data via OPC UA, Modbus or MQTT and sending it to a central system.
- Storage: writing timestamped measurements to a time-series database.
- Dashboard: live machine status, OEE, production rate and defect rate.
- Alerts: notification by email, SMS or push notification when a threshold is crossed.
What is OEE and how is it calculated?
OEE (Overall Equipment Effectiveness) shows, as a single percentage, how efficiently a machine or line uses its planned production time. The formula multiplies three components: OEE = Availability × Performance × Quality. Availability measures how much of the planned time the machine was actually running; performance, how close it came to its ideal speed while running; and quality, how many of the parts produced were good the first time. An OEE of 100% means production ran for the entire planned time, at ideal speed and without a single defective part. Which component is falling short matters as much as the total. Low availability points to breakdowns, waiting for material and long setup times; low performance to slow running and short stops; and low quality to scrap and rework. That is why OEE serves as a compass that shows where to start improving. An accurate calculation requires data collected directly from the machine rather than entered by hand.
| Component | What does it measure? | How is it calculated? |
|---|---|---|
| Availability | How much of the planned time the machine was running | Run time ÷ Planned production time |
| Performance | How close it came to ideal speed while running | (Total count × Ideal cycle time) ÷ Run time |
| Quality | How many parts were good the first time | Good count ÷ Total count |
For example, take a single shift with entirely hypothetical figures:
- Planned production time is 450 minutes (an 8-hour shift minus a 30-minute planned break). Breakdowns and setup cause 45 minutes of downtime, so run time is 405 minutes. Availability = 405 ÷ 450 = 90%.
- The ideal cycle time is 0.6 minutes (36 seconds) per part, and 540 parts are produced during the shift. Performance = (540 × 0.6) ÷ 405 = 324 ÷ 405 = 80%.
- Of the 540 parts, 27 are defective and 513 are good. Quality = 513 ÷ 540 = 95%.
- OEE = 0.90 × 0.80 × 0.95 = 0.684, or roughly 68%.
In this example the biggest loss is on the performance side: the machine is running, but below its ideal speed. The first things to look at are short stops and slowed-down cycles. The figures were chosen only to illustrate the calculation; they do not belong to any plant or customer.
What do anomaly detection and predictive maintenance deliver?
Anomaly detection means automatically catching situations in machine data that deviate from normal behavior: a motor's current slowly creeping up, a bearing's vibration changing, or the cycle time getting longer on a particular product. These deviations often show up in the data days or even weeks before a failure, but are hard to spot by eye. Detection combines statistical methods and machine learning models: the system first learns the machine's normal behavior, then flags measurements that drift significantly from it. Predictive maintenance links these alerts to the maintenance plan. Compared with reactive maintenance, which steps in after a failure, and periodic maintenance, which replaces parts on a schedule, it times maintenance according to the actual condition of the equipment. The main goals are to reduce unplanned downtime, have spare parts ready exactly when they are needed and avoid replacing parts that are still sound too early. A reliable model needs a few months of clean data and maintenance records kept in digital form, so the earlier data collection starts, the better.
| Approach | When is maintenance done? | Weakness |
|---|---|---|
| Reactive | After a failure | Unplanned downtime and urgent need for parts |
| Periodic | By calendar or operating hours | Healthy parts are also replaced early |
| Predictive | When the data shows signs of deterioration | Requires data collection and model setup |
What does BYT Digital do in this field?
BYT Digital is a boutique technology studio founded at Teknopark İzmir that builds industrial data analytics end to end: from sensors and PLCs to the database, and on to the dashboard and alerting system. We collect real-time data from the PLCs and sensors on your production line using OPC UA, Modbus and MQTT, store it in time-series databases and build OEE calculation, anomaly detection and predictive maintenance layers on top. We have experience in the laser cutting, food and textile sectors. So that data feeds into business processes instead of just sitting on a screen, the project can also include ERP and accounting integration; we have worked with SAP, Logo, Mikro and Netsis for ERP, and Paraşüt, Bizim Hesap and e-Fatura (e-invoicing) for accounting. We use this approach live in Oktapus, our own product for the laser cutting sector. Before every integration we review compatibility and risks with your existing machines and software, and make a clear recommendation. You work directly with the founding engineer, with no outsourcing chain in between.
- Real-time data collection from machines, PLCs and sensors (OPC UA, Modbus, MQTT)
- Time-series database and dashboard
- Tracking of OEE, production rate and defect rate
- Anomaly detection and predictive maintenance
- Alerts by email, SMS or push notification
- Integration with ERP and accounting systems
How does an industrial data project run?
The safest way to start an industrial data project is with a single machine or line: validate the data flow and indicators there, then roll out to other machines. At BYT Digital the process has six steps. During discovery (1–2 weeks) together we map out your machines, your existing software and the indicators you want to track; the first meeting and pre-analysis are free. By the end of planning and design (1–3 weeks in total) you have the roadmap, technical architecture, timeline and price in writing. During development we work in short cycles and show a working version after each. Testing and delivery (1–2 weeks) covers automated tests, manual quality checks, documentation and user training. A comprehensive industrial platform typically takes 4–6 months. Bug fixes are free for 30 days after launch; after that, maintenance continues under a monthly service level agreement (SLA), and 24/7 response for critical infrastructure is available under a separate contract. See the project process section on our home page for each step in detail.
Related product: Oktapus
Oktapus is an end-to-end industrial platform that BYT Digital developed for laser cutting shops. It collects real-time data from the machines, prepares automatic pricing and quotes, and brings press-brake bending planning and ERP integration together in a single panel. It runs the shop from order intake to shipment; OEE calculation and predictive maintenance are also part of the platform. It is designed for facilities with 3 to 50 machines. You can explore it as a live, sector-specific example of the data flow described on this page.
Frequently asked questions
Do we need to buy new sensors to measure OEE?
Not always. Basic information such as running status and part counts is already available in the PLC of most machines. Sensors are added for any measurements that are missing. During discovery we determine, machine by machine, what each one needs.
Can data be collected from older machines that have no PLC?
In most cases, yes. Current, vibration or counter sensors can track whether a machine is running and how many parts it produces. This method does not provide detailed process data, but it is usually enough to calculate availability and performance.
Which protocols and systems do you work with?
We have experience with OPC UA, Modbus and MQTT on the machine side; SAP, Logo, Mikro and Netsis for ERP; and Paraşüt, Bizim Hesap and e-Fatura for accounting. We carry out a compatibility and risk analysis before every integration.
What is a good OEE score?
The benchmark most often cited in the literature is 85%, but a realistic target varies with the sector, the product mix and how the line is set up. The most useful comparison is your own line's trend over time and which component the losses are concentrated in.
How long does a project take?
It depends on the scope. Discovery takes 1–2 weeks, planning and design 1–3 weeks in total, and testing and delivery 1–2 weeks; a comprehensive industrial platform typically takes 4–6 months. Bug fixes are free for the first 30 days after launch.