A motor is shaking more than before. Its temperature is rising slowly, but the production process was already running earlier. Will the engineering crew investigate earlier, or will this wait until they are due for a regular equipment check-up?
In a connected, modern auto factory, sensor insights and artificial intelligence can help.
IoT sensors gather physical device data, and AI can help explore the numbers for patterns and insights.
When paired together, the connected car manufacturing ecosystem can help producers make more educated decisions about maintenance, quality inspection, and production process changes.
1. Gathering Necessary Information With IoT
Before making any inferences about a process, an AI engine would require a substantial amount of relevant information.
Industrial IoT platforms can help monitor equipment and the production process by gathering and analyzing relevant insights. Sensors and monitoring services can capture a wide variety of relevant data for automotive manufacturers, including:
• Temperatures and vibrations
• Rotational speed and run-time of a motor
• Cycle durations of manufacturing equipment
• Operational and power status of various devices
• Position and inventory of parts
• Environmental conditions of an area in proximity of manufacturing and testing equipment
For example, a monitoring service could capture temperature and vibrational data of a motor in an automotive assembly plant. These two factors will provide the engineering crew and AI systems with some information about a motor but, typically, not enough. Engineers would have to evaluate how the motor behaves under various conditions before they can predict how the motor will respond at a given moment. For example, engineers would have to build a reference from other similar devices to establish what a “normal” set of vibrations for a specific motor should be.
2. Generating Insights With AI Processing
IoT can enable manufacturers to observe their process. But having raw data about equipment conditions is not enough. To get to meaningful insights, automotive producers would be better off pairing IoT data with artificial intelligence for analysis. There are several ways AI and machine learning can become helpful for manufacturers. A few key use cases include:
Predictive maintenance
By capturing equipment data, a learning model can find behavioral patterns that may signal deterioration in quality. This kind of information can help producers find and address any developing issues.
Automated quality inspection systems
Computer vision enables an ML model to look at images of products and find any problems or irregularities of interest for the manufacturing process. This use case can reduce human effort during the quality inspection phase of production.
Production optimization
By analyzing data from cycle time durations, production equipment status, and interruption frequencies and causes, AI can help optimize a manufacturing process. This data can provide valuable insight into what manufacturing stages and equipment should be prioritized.
Anomaly detection in IoT data
Detecting unusual patterns of behavior can also be achieved by creating a reference of what a “normal” observation should be and flagging any outliers. These irregular patterns can include any information that deviates from established behavioral reference.
The choice of ML technique that should be leveraged for a specific use case will depend on the particular needs, volumes of available information, and acceptable prediction risks.
3. An Illustrative Predictive Maintenance Workflow
Let’s imagine a motor that runs continuously in an automotive manufacturing plant. One of the initial solutions that come to mind in this situation is a predictive maintenance approach.
A monitoring workflow that incorporates AI inference may have the following steps:
Step 1: Sensors
collect vibrational and temperature data at regular intervals
Step 2: Data pipelines
validate this information, verifying that all equipment has proper timestamps and that no values are empty or malformed.
Step 3: Engineers
establish the behavior of this motor at particular conditions, such as higher or lower loads.
**Step 4: **Either a model or an anomaly detection solution will compare new data with the information from step three or the established ranges for this motor.
Based on this data, an engineer would choose either a statistical or a machine learning model for this AI application or another approach (e.g., time series), depending on the volume of available data, the frequency and predictability of anomalous patterns, and other factors.
**Step 5: **Once the engine detects an outlier, it could issue an alert that will prompt human judgment
about the technical condition of the motor. This alert would usually include contextual information about the event, such as timestamps, equipment identifiers, and, possibly, other relevant variables.
In step 6, once there is a human evaluation of the equipment’s current state, engineers will make a conclusion and decide whether it is necessary to halt the production process and schedule maintenance for this equipment.
As you can see, the AI process is not a deterministic one an alert will usually provide valuable information, but not one of a guaranteed equipment or process failure. An anomaly detection model flags values that fall outside of expected ranges.
A predictive maintenance model that flags the likelihood that an equipment may fail at a later point in time involves a few additional steps because it needs to analyze what anomalous patterns may represent equipment wear that could lead to an imminent failure.
However, if the alert system is configured incorrectly, the model could raise false alarms frequently. This makes evaluation of the model’s performance against false positives an equally valuable metric during the solution evaluation phase.
4. Assessing Real-World Performance of an AI Solution
The actual evaluation of any applied model in the physical world is also far from complete. Engineers and analysts will want to measure the following performance indicators specific to a predictive maintenance use case:
1. False alarm rate: How often the maintenance alert system falsely marks equipment as faulty.
2. Missed fault rate: How often such an AI system fails to detect an equipment or process issue.
3. Detection lead time: A crucial time indicator that suggests how much time there will most be be available ahead of a failure for maintenance crews to investigate.
4. Alert response: How soon can the engineering team react upon receiving a model alert.
5. Process interruptions: Will this maintenance platform prevent unplanned system downtimes?
6. Sufficiency of data: Is the volume of incoming data adequate for this model?
Additionally, it will be important to balance how impactful false positives and false negatives will be for the production process. Failing to detect serious equipment issues will likely have negative consequences similar to unplanned downtimes during production, but these consequences will also vary from one setup to another.
Although a model will initially show promising results during training, it will also be crucial to continually monitor the platform and reevaluate its performance after changes, such as new equipment being added or different workload patterns.
5. Expanding Manufacturing Applications to Include Car Safety Features
The same kinds of broad principles that connect IoT devices and AI models apply to modern car safety technology. Automotive manufacturers can install onboard cameras, radar systems, LiDAR, and other sensors to assess their surrounding environment.
AI powered perception could help evaluate the camera feed from these sensors, giving insight into the car’s surroundings and enabling features such as lane-keeping and automatic brake mechanisms. Of course, actual vehicle safety extends beyond model inference to the design and installation of appropriate onboard sensors, systems validation, cyber physical security, and a variety of other factors. Actual automotive safety evaluations will usually require testing the product in relevant, varied, and realistic conditions.
6. Merging AI Components With Existing Factory Infrastructure
Modern auto factories are known to operate multiple systems that control, monitor, and optimize various processes during production. Different solutions serve various purposes for manufacturers:
• Manufacturing execution systems manage and control the production process.
• Enterprise resource planning tools organize and strategize business operations.
• Quality management systems collect and organize quality data and information about manufacturing anomalies.
• Equipment monitoring platforms gather and store data about individual equipment.
When these tools are standalone solutions, it may be difficult for engineers to investigate any production issues. For example, combining quality inspection data about an issue with the relevant machine data and production timeline is essential when looking for the root cause of any manufacturing discrepancy. To optimize the manufacturing process, engineers may want to analyze equipment, part, production scheduling, and maintenance data.
In an AI connected system, these pieces of information will need to be integrated into one unified solution, and engineers may need to consider the following:
• Communication standards: Most factory devices will communicate using standard sets of protocols, but manufacturers may need to build or choose additional communication infrastructure when these standards cannot be met.
• Application programming interfaces (APIs): Where these are available and compatible between systems.
• System availability: Will the necessary systems be up and running at all times to supply the necessary data?
**• Edge computing: **Edge services usually deliver low-latency performance and processing to individual manufacturing devices when necessary. On the other hand, a central cloud platform can be helpful to unite disparate systems, ensure sufficient data storage capacity, and provide the overall flexibility necessary for an AI solution. The particular choice of a design philosophy will depend on specific infrastructure requirements.
7. Additional Manufacturing Challenges Beyond the Core AI Engine
Even the most carefully designed and well implemented AI system will often under-perform if the supporting platform has issues.
Poor data collection and storage: When essential information is missing or inconsistent.
Legacy equipment: Older devices or machinery at a manufacturing site that will require additional infrastructure to support IoT solutions.
Variations in workload: An AI model can become less reliable and accurate when workload requirements change significantly.
**Cybersecurity: **Connecting a factory monitoring platform requires cybersecurity and network segmentation infrastructure to protect devices, data transmission, and processing services from threats.
Human validation: An alert about maintenance may need to be reviewed and corroborated by a human worker or team, depending on the risk level and technical requirements of a given scenario.
System design: Network downtime or unavailable hardware and software can severely impede the ability of an AI-driven factory monitoring solution to deliver on its promised performance. These weaknesses in a broader system should be identified and addressed prior to launch rather than at a later date.
Conclusion
AI and IoT can help manufacturing producers turn equipment and production process data into more valuable and actionable insights. IoT will enable visibility, and AI will help find meaningful patterns that can aid the maintenance process, quality inspection, and production cycle optimization. The best approaches will begin with a well defined objective, collect relevant data, and consider how AI alerts will be routed towards maintenance, inspection, and other operational teams. For readers interested in exploring how connected technology is applied in automotive manufacturing,(https://oemnexai.com/) is a good resource. A reliable AI solution in manufacturing does not always begin with a deep learning model, but rather with a focus on helping the appropriate people get reliable, timely insights to help them take timely and accurate action.
Top comments (0)