Anomaly Detection in Plastics Manufacturing
We operate in an industry where economic pressures, fierce competition, and ever-changing consumer demand challenge plastics manufacturers every day. Despite all this unpredictability, our industry is still expected to run cost-effective, high-quality, and efficient production processes.
Why Material Behavior Matters
From our perspective, the key to a cost-effective process lies in understanding the numbers. In plastics manufacturing, anywhere from 50% to 70% of the final product’s cost can be attributed to materials. To minimize waste and maintain efficiency, it’s crucial to gain deep, real-time insight into material behavior.
Why?
Because seemingly small details like storage conditions, in-mold temperatures and pressures, or batch-to-batch deviations can drastically affect how materials behave during processing. Relying solely on standard material data sheets or simulations risks overlooking these variations, ultimately leading to unexpected quality issues.
The Industrial Status Quo
Many manufacturers recognize that material parameters can fluctuate for a variety of reasons, thus employing different strategies to maintain quality. One common method involves manual adjustments, where operators rely on a mix of expertise and periodic checks to fine-tune process settings. While this approach can certainly address basic shifts, it often lacks the speed and precision required to respond effectively to real-time changes in material behavior.
Other widely used resources are standard data sheets and simulations, which can be invaluable when setting up initial process parameters. However, these documents and models may become less reliable as actual production conditions begin to deviate from the presumed norms. They also do not account for the subtle day-to-day variations in material properties or in the manufacturing environment itself, potentially leading to overlooked deviations and quality issues.
Anomaly Detection as a Solution
Having explored the industrial status quo in addressing material variability, it’s evident that truly effective quality control requires a deeper, more responsive strategy. This is where anomaly detection in manufacturing comes into play.
What is anomaly detection?
Anomaly detection is a method of data analysis used to distinguish patterns within a dataset and determine data points that fall outside the identified patterns. In simpler terms, anomaly detection is a way of uncovering outliers from average values.
Anomaly detection is a classic example of the unsupervised learning method in machine learning, a subset of artificial intelligence (AI). In unsupervised learning, an algorithm uncovers patterns and outliers within a dataset without the aid of predefined labels or correct outcomes.
How is anomaly detection typically used in plastics manufacturing?
In plastics manufacturing, anomaly detection can help identify unexpected variations in production, such as process drifts or equipment wear. Typically, anomaly detection is used to increase transparency by tracking machine and sensor data, identifying deviations, and implementing some form of quality control during the production process.
However, anomaly detection is not enough if we consider the challenge and importance of understanding real-time material behavior to control process outcomes and remain cost-effective.
Enhancing Process Transparency Through Anomaly Detection and Material Characterization
By collecting material insights through our advanced dielectric sensor technology, sensXPERT develops powerful machine learning algorithms. While these models are in development, sensXPERT can already provide manufacturers with valuable process insights from the start of its solution’s integration.
These process insights can come in the form of straightforward anomaly detection – i.e., tracking incoming data, monitoring deviations in the data, and facilitating basic quality control. Additionally, basic anomaly detection can be combined with the material behavior data collected by the dielectric sensors for greater transparency into what the detected anomalies are.
After building a robust machine learning model that generates holistic information and predicts material behavior progressions throughout the process, sensXPERT can still implement anomaly detection to draw conclusions about the process as a whole.
Think machine data and machine parameters. sensXPERT uses OPC UA for data transfer with machines and existing supplier systems, taking into account all parameters that influence the process. The process control technology can thus create significant benefit for manufacturers by identifying anomalies within these parameters and any additional incoming data. Therefore, should a machine present a negative trend indicating it might face issues or need maintenance, sensXPERT can flag it in advance and help manufacturers prevent excessive downtime.
Practical Use Cases
Our anomaly detection approach becomes clear when looking at two specific examples. One involves monitoring the adhesion between two different materials in an in-mold coating process, while the other focuses on maintaining quality in an e-mobility application prior to assembly.
Refining In-Mold Coating
In the in-mold coating scenario, our goal was to enhance product quality by fine-tuning machine parameters. Although operators frequently made minor adjustments, the actual impact on product consistency remained uncertain. The main challenge lay in determining which parameter changes could negatively influence final part quality. To tackle this, we analyzed the manufacturer’s machine performance data across 17 distinct parameter settings.
We then applied our anomaly detection throughout the production cycle to identify parameter drifts. Anomaly detection identified each machine parameter rate with a false positive rate of 0.02%, meaning that each parameter drift was detected and only 0.02% of the flagged anomalies were not a change in machine parameters.
In every instance of detected drift, sensXPERT empowered the processor to validate final part quality and adjust the relevant parameters. This dynamic approach ensured a more consistent outcome in the end product.
Ensuring Quality in E-Mobility
In another example, we supported an e-mobility manufacturer seeking to identify potential quality issues associated with introducing a new material in production.
By employing our anomaly detection, we pinpointed marginal process areas where deviations were occurring. With these anomalies flagged, the manufacturer could swiftly locate and inspect any parts deemed suspicious.
This process made it possible to label critical data points, which we then fed into our advanced machine learning algorithm to bolster defect detection. Over time, these insights pave the way for more dynamic adjustments to process parameters, further optimizing production efficiency.
By pairing anomaly detection with real-time material insights, we at sensXPERT provide the level of clarity and control required to produce top-notch products while still managing costs effectively. In an industry challenged by limited transparency, this powerful combination enables manufacturers to remain agile, competitive, and prepared for whatever lies ahead.
Learn More About Anomaly Detection in Plastics Manufacturing
