Plastic injection molding is one of the most widely used manufacturing processes in the world, producing everything from automotive and electronic components to medical devices, packaging, and everyday consumer products. And the market continues to grow. The global injection-molded plastics market was valued at approximately $362.5 billion in 2025 and is projected to reach $481.4 billion by 2033, driven in part by increasing demand for high-volume, cost-effective plastic components. At the same time, the global market for plastic injection-molding machines is expected to expand as manufacturers increase production across automotive, packaging, electronics, healthcare, and consumer-goods applications.
The reason for injection molding’s widespread adoption is straightforward: it can manufacture complex parts quickly, repeatedly, and at enormous scale. But producing millions of parts also means that even a relatively small quality problem can quickly translate into significant scrap, rework, downtime, and cost.
The challenge is growing as products become smaller, lighter, thinner, and more complex, while dimensional tolerances tighten. Manufacturers are expected to maintain consistent quality even as material characteristics, temperatures, pressures, machine conditions, and other process variables change from one production run to another, or even within a single molding cycle.
Traditionally, maintaining this consistency has depended heavily on experienced operators, established process limits, and inspection of finished parts. But by the time conventional inspection identifies a problem, the process may already have produced dozens or hundreds of defective parts.
What if we could predict the quality of the part before inspection tells us there is a problem?
That is where machine learning becomes particularly interesting.
An injection-molding machine generates a tremendous amount of process data during every cycle: injection pressure, holding pressure, melt temperature, mold temperature, cavity pressure, cooling time, screw position, cycle time, and more. Rather than simply displaying or storing this data, machine learning can use the relationships hidden within it to predict the quality of the finished product.
And this is where three technologies can work particularly well together:
Regression: What will the finished part’s quality measurement be?
Binary Classification: Will the finished part pass or fail inspection?
Rule-Based Expert System: Based on those predictions and our engineering knowledge, should the process continue, be adjusted, or be stopped?
Together, these capabilities can move quality control from detecting defects after they occur toward predicting quality while the product is being manufactured, and providing operators with actionable guidance before a small process deviation becomes a much larger production problem.
From Reactive Quality Control to Predictive Quality
Maintaining consistent quality in injection molding can be challenging. Small variations in material properties, temperature, pressure, injection speed, cooling time, and mold conditions can contribute to defects such as warpage, sink marks, flash, short shots, and dimensional variations, resulting in scrap, rework, lost production time, and higher manufacturing costs.
These challenges become even more significant as manufacturers produce increasingly complex parts with tighter tolerances while responding to changes in materials, equipment conditions, production volumes, and customer requirements.
This creates an opportunity to move from simply detecting quality problems to predicting them.
Instead of relying solely on finished-part inspection, manufacturers can use the process data generated during each molding cycle to identify patterns associated with product quality and potentially recognize developing problems earlier.
Turning Process Data into Predictive Quality
An injection-molding machine already generates a tremendous amount of data during every production cycle. Variables such as injection pressure, holding pressure, melt temperature, mold temperature, cavity pressure, screw position, cooling time, and cycle time provide a detailed picture of the conditions under which each part was produced.
Traditionally, much of this information has been used for monitoring, alarming, trending, and troubleshooting. But what if we could use the same data to predict the part’s quality as it is being produced?
By combining historical process data with actual inspection and quality results, machine-learning models can learn relationships between process conditions and finished-part quality. However, machine learning does not have to work alone. Engineering knowledge and established process limits can provide another important layer of protection.
As illustrated in the diagram, our predictive-quality example uses four layers:
Let us follow one injection-molding cycle through the system.
Layer 1: Rule-Based Gatekeeper — Is the Process Operating Within Known Limits?
Before asking machine learning to predict quality, the first layer evaluates conditions that engineers already understand.
A Rule-Based Expert System can act as an immediate gatekeeper by comparing machine-cycle data against established process limits and checking for known machine, sensor, or process faults.
Rules use straightforward IF–THEN logic.
For example:
IF barrel temperature exceeds the validated upper limit,
Then flag the process and alert the operator.
Or:
If cushion position falls below the validated minimum,
Then identify the cycle as abnormal and hold the part for inspection.
These rules are not intended to replace machine learning. They capture conditions where the manufacturer already knows what constitutes an unacceptable or abnormal process state.
If a critical condition is detected, the rule-based system can immediately generate the appropriate response. If the process passes these initial checks, the cycle data moves to the machine-learning models for deeper analysis.
Layer 2: Regression — What Will the Quality Measurement Be?
The next layer uses Regression to predict a continuous numerical value from the process data.
In injection molding, regression can be used to predict measurable quality characteristics such as:
Part weight— predicts mass consistency based on process conditions such as packing pressure, temperature, and filling behavior.
Critical dimensions— estimates dimensions that may be affected by pressure, temperature, packing, and cooling conditions.
Warpage— predicts the amount of bending or deformation in the finished part.
Shrinkage— estimates dimensional change as the polymer cools and solidifies.
Other measurable quality characteristics— depending on the application and available training data, additional physical or mechanical properties may also be predicted.
For example, assume the manufacturer has established an acceptable part-weight specification of:
84.5–85.5 g
The regression model evaluates the current molding-cycle data and predicts:
Predicted Part Weight: 84.7 g
The predicted weight is within specification.
But that does not necessarily mean that everything about the part is acceptable.
A part could have the correct weight and still contain a quality defect. This is where the next layer becomes important.
Layer 3: Binary Classification — Will the Part Pass or Fail?
Binary Classification answers a different question.
Instead of predicting a measurement, it predicts one of two outcomes:
PASS or FAIL
The classification model is trained on historical molding cycles, in which process data can be associated with the actual inspection result for the finished part. It learns patterns and relationships across multiple variables that may distinguish acceptable parts from defective ones.
For the current cycle, suppose the model produces:
Predicted Result: PASS
Probability of Failure: 6%
Now we have two complementary pieces of information.
The regression model tells us:
The predicted part weight is 84.7 g and is within specification.
The classification model tells us:
The part is predicted to pass, with only a 6% probability of failure.
Regression answers:
What will the quality measurement be?
Binary Classification answers:
Will the finished part pass or fail?
Together, they provide a much richer understanding of the part’s expected quality than either model could alone.
But there is still one critical question:
What should we do with those predictions?
For the current cycle, suppose the model produces:
Predicted Result: PASS
Probability of Failure: 6%
Now we have two complementary pieces of information.
Regression answers:
What will the quality measurement be?
Binary Classification answers:
Will the finished part pass or fail?
Together, they provide a much richer understanding of the part’s expected quality than either model could alone.
But there is still one critical question:
What should we do with those predictions?
Layer 4: Rule-Based Decision — What Should We Do Next?
This is where the Rule-Based Expert System can be used again.
Instead of acting as the initial gatekeeper, it now acts as a decision layer, combining:
– Regression predictions
– Binary Classification results
– Prediction probabilities
– Validated process limits
– Current process conditions
– Consecutive-cycle behavior
– Engineering knowledge
– Quality procedures
The expert system can then determine the most appropriate response.
For example:
IF predicted part weight is within specification
AND probability of failure is low
AND no critical process rules have been triggered
THEN continue production.
But consider a different cycle:
IF predicted part weight is within specification
AND probability of failure is increasing
THEN continue with increased monitoring and inspect the next part.
Or a more serious condition:
IF predicted part weight is outside specification
AND probability of failure exceeds the established threshold
THEN hold the part and recommend that the operator investigate the process.
And if the problem persists:
IF probability of failure remains above the critical threshold for multiple consecutive cycles,
THEN stop production and notify the operator to investigate.
The exact limits and actions would be determined by the manufacturer based on the validated molding process, quality requirements, and engineering knowledge.
The Rule-Based Expert System can go a step further by helping the operator determine how to respond to the problem. Rather than simply identifying an abnormal condition or recommending that the process be stopped or adjusted, the system can provide step-by-step troubleshooting guidance based on established engineering knowledge. For example, it could recommend which process parameters to check, suggest potential causes, and guide the operator through the appropriate corrective actions.
The objective is not simply to generate another prediction for an operator to interpret. The objective is to turn process data and machine-learning results into clear, explainable, and actionable information.
Moving from Reactive to Predictive Quality
The real value of this architecture lies in bringing together two distinct types of intelligence.
Machine learning learns from the data.
Regression and Binary Classification can discover relationships across process variables and historical quality results that may be difficult to capture with conventional limits alone.
The Rule-Based Expert System captures engineering knowledge.
It provides deterministic rules for known conditions and helps translate machine-learning predictions into actions that operators can understand.
Together, they create a progression from:
Process Data → Prediction → Engineering Knowledge → Decision → Action
Instead of discovering a quality problem only after finished-part inspection, manufacturers can begin identifying developing conditions during the molding cycle, predicting the likely quality of the finished product, and giving operators the information they need to respond earlier.
The goal is not to replace the process engineer or operator. It is to give them better information, earlier in the process.
Predict what is likely to happen. Determine whether it matters. Apply engineering knowledge. Then take the appropriate action.
From Prediction to Better Quality
Predictive quality control represents an important shift in manufacturing, from detecting defects after they occur to identifying potential quality problems before they result in scrap, rework, or lost production. By applying machine learning to the process data manufacturers are already collecting, organizations can uncover patterns, predict quality outcomes, and give operators earlier insight into developing problems.
But prediction alone is not enough.
As we demonstrated with the injection-molding example, the real value comes from combining machine learning with engineering knowledge. Regression can predict what a finished-part measurement is likely to be. Binary Classification can estimate whether the part will pass or fail inspection. A Rule-Based Expert System can evaluate those predictions within the context of established process limits and manufacturing expertise, and then recommend what the operator should do next.
The result is a progression from:
Process Data → Prediction → Engineering Knowledge → Decision → Action
The Rule-Based Expert System can also take this one step further by helping operators troubleshoot a problem. Instead of simply displaying an alarm or recommending an adjustment, rules can identify potential causes and guide which process parameters to investigate and what corrective actions to consider.
Machine learning predicts. Engineering knowledge guides the decision. The operator remains informed and in control.
Ready to Explore Machine Learning in HMI/SCADA?
Attend the next webinar, Machine Learning for Industrial Automation: Solving Real-World Problems with ADISRA SmartView 5.0, on:
August 19 — Portuguese
August 20 — English
See the webinar information below for additional details and the registration link.
Beginning September 3, you can download a trial version of ADISRA SmartView 5.0 and explore how machine learning and a Rule-Based Expert System can be incorporated directly into an HMI/SCADA application. Download ADISRA SmartView V5.0 here.
If you would prefer a private demonstration of ADISRA SmartView 5.0, click on this link and complete the request form.
Move beyond simply monitoring your process. Start using your industrial data to predict, understand, and improve what happens next.
Machine Learning for Industrial Automation: Solving Real-World Problems with ADISRA SmartView 5.0
Everyone is talking about analytics and artificial intelligence (AI), but how can machine learning actually improve industrial operations?
With ADISRA SmartView 5.0, machine learning is built directly into the HMI/SCADA development environment and works alongside an integrated rule-based expert system. This powerful combination enables engineers to develop applications that not only monitor operations but also predict issues, classify conditions, detect anomalies, and recommend actions.
In this webinar, you’ll learn:
– Which machine learning algorithm should I use?
– What industrial problems does each algorithm solve?
– How do I train a machine learning model?
– How do I integrate the results into my HMI/SCADA application?
This webinar series answers these questions by exploring one machine learning algorithm at a time, demonstrating how it can be applied to solve real-world industrial challenges. This session includes practical examples that you can apply immediately to your own automation projects.
Join us:
– Portuguese: August 19, 2026, at 8:30 AM CDT / 10:30 AM BRT
– English: August 20, 2026, at **9:30 AM CDT
Whether you are developing new automation systems or enhancing existing applications, you will gain practical insights into how to use machine learning to create smarter, more intelligent HMI/SCADA solutions.
Register today and discover how built-in machine learning can transform your industrial automation applications with ADISRA SmartView 5.0. Register here.
ADISRA®, ADISRA’s logo, InsightView®, KnowledgeView® and RapidView™ are registered trademarks and trademarks of ADISRA, LLC.