Artificial Intelligence (AI) is emerging as a key player in manufacturing, logistics, warehousing, and supply chain functions. AI is being utilized by businesses for demand forecasting, equipment failure prediction, inventory optimization, anomaly detection, and production planning.
AI, however, is not as powerful as the data it’s fed. In a production center, Radio Frequency Identification (RFID) can provide real-time information about the products, components, equipment, containers, and materials. Even the most advanced AI algorithms can generate erroneous conclusions if the information extracted from the RFID systems is incomplete, duplicated, inaccurate, or poorly organized.
This makes RFID data quality a foundation for successful industrial AI.
The Relevance of RFID Data Quality
RFID can be used to identify and track objects’ movement automatically without having to scan all objects by hand. However, an RFID reader is not able to interpret the meaning behind each and every signal in the context of the business.
A system can store the information that a tag was detected at a specific reader at a specific time. The business must decide if that event is an indication that the item has been received, moved, stored, processed, shipped, or if it was only re-detected since it was still within range of the reader.
Thus, poor-quality data can be used to draw erroneous conclusions.
For reliable data useful for AI applications, it should be:
- Accurate: It depicts the actual event in the real world accurately.
- Complete: There are no missing important movements & records.
- Consistent: Data is structured consistently and uses consistent rules.
- Timely: Information arrives to applications in a timely fashion, allowing for timely decisions.
- Only useful events are passed to analytical systems
Common Sources of Poor RFID Data
There are a variety of reasons that impact the quality of the data in an industrial RFID environment.
1. Missed Reads
Readers may miss tags due to metal, liquids, tag orientation, interference, distance, and inappropriate tag selection. If an AI system doesn’t receive a read, it may think that an item never arrived at a specific location.
2. Duplicate Reads
An RFID reader can read a tag multiple times if the tag is in the reader’s field of view. If data isn’t filtered, thousands of repeat reads will clobber databases and skew analytics.
3. False Reads
Sometimes readers can be aware of tags being outside of the location where they were intended. If a reader is not placed properly or if the coverage area overlaps, it may be assumed that there is inventory or equipment in an area where there is not.
4. Incorrect Tag Associations
When a tag is used on the wrong Product, Container, or Work Order, then all subsequent events will be tied to incorrect information.
5. Inconsistent Data
Production lines or facilities could have different naming conventions, timestamps, identifiers, and/or event definitions. AI models can have difficulties when the data they are supposed to be the same is formatted differently.
RFID Reads to Meaningful Events Converter
The first and foremost step is to differentiate business events from raw RFID reads.
A reader, for instance, could read the same pallet 50 times as it flows through a dock door. It’s not required that the AI system have all 50 detections. It may require one validated event where the pallet has entered the receiving area.
An event-processing layer or RFID middleware may be of assistance:
- Filter duplicate reads
- Apply time windows
- Match the names of readers to where they are. Match the names of readers with where they are.
- Detect movement patterns
- Validate expected sequences
- Create common business events
This converts the radio-frequency noise into information for use by AI systems.
RFID and Predictive AI
Good RFID information can be used in various industrial AI applications.
Manufacturers can, for instance, use the RFID movement history in conjunction with production information to determine if there are any unusual delays. AI tools could identify that a component is typically moving from one workstation to the next in 30 minutes, but takes multiple hours to do so over a period of time, and flag this as a potential bottleneck.
Likewise, the historical movement data that RFID can offer can assist AI to identify:
- Unusual inventory movements
- Production delays
- Recurring process bottlenecks
- Abnormal equipment utilization
- Inefficient material flows
- Potential inventory discrepancies
What’s crucial is that AI does not build reliable intelligence out of unreliable observations.
Integrating RFID into Other Industrial Data
RFID is even more powerful when coupled with other data sources.
A manufacturer may be able to link RFID information to:
- Machine sensors
- MES data
- ERP records
- Warehouse management systems
- Environmental sensors
- Quality-control information
- Maintenance records
- Production schedules
For example, RFID could tell that a component was delivered to a machine, whereas the sensor data tells that the machine was operating under an abnormal condition. AI can combine both sets of data to uncover relationships where humans wouldn’t see them.
Establishing Data Governance
Industrial organizations need to have policies in place for the control of RFID information.
Data governance should establish:
- Who owns RFID data
- System of tagging and tag maintenance.
- The time that information can be remembered
- How duplicate reads are processed
- The methods used to investigate exceptions
- When are events deemed to be authoritative?
The way data is harmonized between facilities. The manner of data standardization between facilities.
As the number of RFID deployments grows from single production lines to factories/supply networks, these processes become more relevant.
The Following Tips Will Help in enhancing the Quality of the RFID data:
Several practical steps can help to enhance the RFID foundation of businesses:
- Test tags in real environments: Assess performance of real products, packaging, machinery, and containers.
- Design read zones with care; do not depend on maximum reader power.
- Detect repetitive reads: Turn repeated detections into meaningful movement events.
- Establish consistent IDs for tags, products, locations, and equipment.
- Track missed, duplicate, and unexpected reads – monitor read performance.
- Check data against reality: Check RFID data against reality from time to time, or compare with manual counts or known movements.
- Define the criteria for receiving, moving, processing, and shipping an event.
- Take the baby steps: Ensure data quality in a pilot project before expanding to the rest of the organization.
- Create the Data Foundation before the AI Layer.
To start an industrial AI project, sometimes the initial starting point is an algorithm, dashboard, or predictive model. But it’s time to pay attention to data infrastructure as well.
Prior to implementing AI, companies should ask themselves:
Is the data from the RFID accurate?
If the result is unclear, better data collection and processing could yield greater returns than accessing a more sophisticated AI model right away.
An AI model that is less complicated and trained on clean and consistent, well-understood data can prove more useful than a more sophisticated model trained on noisy data.
For industrial companies, the benefits of RFID lie in its ability to give them visibility over the movement and status of physical assets. However, its true power is beyond tracking. An accurate, timely, standard, and well-understood set of RFID data can be a valuable input into industrial AI.
FAQ
- Why is RFID data quality important for industrial AI?
Industrial AI is dependent on accurate and reliable data to produce useful models. Inaccurate RFID data, including missing, duplicated, and erroneous readings, may lead to incorrect conclusions and predictions. Quality RFID data may provide more reliable insights into industrial processes for AI compared to lower-quality data.
- How can businesses improve RFID data quality?
Companies can ensure quality RFID data by choosing relevant tags, ensuring optimal reader placement, removing duplicate RFID readings, standardizing RFID identifiers, monitoring RFID performance, and cross-checking RFID data with actual processes. Testing RFID systems in industrial facilities during pilot programs can also help ensure quality data by detecting potential issues ahead of time.
- Can we combine RFID data with other industrial data sources?
Yes. RFID data can be combined with data from MES, ERP systems, machine data, maintenance records, quality control systems, and production schedules. This combination can allow AI to better understand the complete industrial process and detect areas for improvement.
