Glossary - The Terms Behind the Tech
Glossary of AI Terms Relevant to Manufacturers
Here are short definitions of the AI and data terms manufacturing teams encounter most often—especially in quality inspection, predictive maintenance, and dock automation. If something is missing, we’d be happy to add it.
Definition
In manufacturing quality control, accuracy is the ratio of correctly predicted cases (both true positives and true negatives) to all inspected cases.
Why it matters on the shop floor
Accuracy gives a quick overall sense of how well an inspection or classification system is performing, but it has to be read together with recall and specificity to understand whether defects are being missed or good parts are being scrapped.
Definition
Agentic AI refers to AI systems that can take a series of actions toward a goal—such as planning steps, calling tools or software systems, and adapting based on feedback—rather than only answering a single query or making a one-off prediction.
Why it matters on the shop floor
Agentic AI can, in principle, coordinate multi-step tasks such as pulling data from different systems, checking model outputs, drafting reports, or suggesting follow-up actions for engineers, which over time could help automate more of the “glue work” around quality, maintenance, and production decisions—not just the underlying predictions.
(more)
Definition
Anomaly detection is an AI-based method that automatically identifies unusual patterns or outliers in data, such as unexpected vibration, abnormal temperature, or missing parts.
Why it matters on the shop floor
Anomaly detection can flag emerging equipment problems or quality issues early, often before hard limits are exceeded or operators would notice them.
Definition
Artificial intelligence (AI) is a branch of computer science that builds systems that can perform tasks that usually require human intelligence, such as recognizing patterns, learning from data, and making decisions.
Why it matters on the shop floor
AI powers applications like automated visual inspection, dock verification, predictive maintenance, and process optimization, turning existing sensor and image data into concrete decisions and recommendations.
Definition
A classification model is an AI model that assigns each input to one of a set of discrete categories, such as “OK” vs “defective” or multiple defect types.
Why it matters on the shop floor
Most visual inspection and dock-verification systems are classification models, so understanding this term helps explain how the system decides which bucket each product or pallet goes into.
Definition
Cloud computing means using remote data centers, accessed over the internet, for storage, processing, and running applications instead of relying only on local servers or on-premises hardware.
Why it matters on the shop floor
Cloud resources can store large amounts of image and sensor data, run heavy training jobs for AI models, and centralize analytics across plants, while edge devices handle real-time decisions on the line.
Definition
Computer vision is a field of AI that uses images or video to detect, measure, and classify what is happening in a scene, such as identifying objects, reading text, or spotting defects.
Why it matters on the shop floor
Computer vision is the core technology behind automated visual inspection and dock verification—replacing or augmenting human inspection with systems that can watch every product, pallet, or label at line speed.
Computer vision is a field of AI that uses images or video to detect, measure, and classify what is happening in a scene, such as identifying objects, reading text, or spotting defects.
Why it matters on the shop floor
Computer vision is the core technology behind automated visual inspection and in- and outbound inspection/dock verification—replacing or augmenting human inspection with systems that can watch every product, pallet, or label at line speed.
Definition
A confusion matrix is a table that compares a model’s predictions with the true labels, showing how many items were correctly classified and how many fell into each type of error (true positive, false positive, true negative, false negative).
Why it matters on the shop floor
Looking at the confusion matrix helps engineers understand whether an inspection model is mostly missing defects, scrapping good parts, or performing well on both—and guides decisions about retraining and threshold settings.
Definition
A convolutional neural network (CNN) is a type of neural network designed for image and video data that learns to detect features like edges, textures, and shapes and uses them to classify or locate objects.
Why it matters on the shop floor
CNNs are the workhorse behind most modern visual inspection systems, enabling reliable detection of surface defects, missing components, and label issues at line speed.
Definition
Deep learning is a subfield of machine learning that uses neural networks with many layers to learn complex patterns in data such as images, sound, or time-series signals.
Why it matters on the shop floor
Deep learning enables robust visual inspection and predictive models that can handle noisy, variable manufacturing data and still detect subtle issues.
Definition
A decision threshold is the cut-off value a model uses to turn a continuous score (for example, a defect probability between 0 and 1) into a binary decision such as “OK” or “NOK.”
Why it matters on the shop floor
Adjusting the threshold lets engineers trade off missed defects versus false rejects, which directly affects scrap, rework, and customer complaints.
Definition
Descriptive AI (or descriptive analytics) analyzes historical data to summarize what has happened—for example, trends in defect rates, downtime, or process parameters.
Why it matters on the shop floor
Descriptive AI provides the baseline understanding of past performance that teams need before moving to predictive or prescriptive approaches.
Definition
A digital twin is a virtual representation of a physical asset, process, or system that is updated with real data from sensors and equipment.
Why it matters on the shop floor
Digital twins make it possible to simulate changes, test “what-if” scenarios, and monitor equipment or lines in a virtual environment before making changes in production.
Definition
Edge computing means processing data on devices close to where it is generated—such as an industrial PC or embedded controller—rather than sending everything to a distant cloud server.
Why it matters on the shop floor
Running AI at the edge allows visual inspection, dock checks, and equipment monitoring to keep working even if the network is slow or down, and it reduces the bandwidth needed to move high-volume image and sensor data.
(more)
Definition
An edge device is the local hardware (for example, an industrial PC, GPU box, or smart camera) that runs AI models close to where data is generated, such as next to a line or at a dock door.
Why it matters on the shop floor
Edge devices are what actually execute real-time inspection, counting, or monitoring models without depending on a constant, low-latency connection to the cloud.
Definition
Explainable AI (XAI) refers to methods and tools that make AI model decisions easier for humans to understand, for example by showing which inputs influenced a prediction and how.
Why it matters on the shop floor
XAI helps engineers and operators trust and validate AI systems, debug unexpected behavior, and connect model outputs back to familiar process knowledge.
Definition
In manufacturing quality control, a false negative occurs when a defective product is incorrectly classified as good and allowed to pass.
Why it matters on the shop floor
False negatives are critical because they represent defects that escape to downstream processes or customers, potentially causing rework, recalls, or safety issues.
Definition
In manufacturing quality control, a false positive occurs when a good product is incorrectly classified as defective.
Why it matters on the shop floor
False positives drive unnecessary scrap and rework and can slow down lines, so they need to be controlled while still catching real defects.
Definition
A feature is an individual measurable property or input used by a model, such as pixel intensity, vibration level, temperature, or line speed.
Why it matters on the shop floor
Choosing and logging the right features—what sensors to use, what signals to keep—largely determines how well an AI system can learn and diagnose issues.
Definition
Generative AI refers to models that can create new content—such as text, images, code, or synthetic data—rather than only making predictions or classifications.
Why it matters on the shop floor
Generative AI can support tasks like creating documentation, summarizing logs, or generating synthetic defect examples for training vision models, helping engineers and operators work faster with complex information.
Definition
A human-machine interface (HMI) is the screen or control interface through which operators monitor and interact with machines or processes, such as touchscreens or panel displays.
Why it matters on the shop floor
Clear, well-designed HMIs help operators understand what is happening, respond to alarms, and adjust parameters quickly and safely.
Definition
HMI simplification means redesigning interfaces so that operators see only the most important information and controls, with complex settings grouped or automated behind the scenes.
Why it matters on the shop floor
Simplified HMIs can reduce setup time, lower training requirements, and reduce the risk of operator error, especially on complex equipment.
Definition
A human-in-the-loop system is one where humans review, correct, or approve AI outputs rather than letting the system act fully autonomously.
Why it matters on the shop floor
Keeping humans in the loop allows experienced operators and engineers to catch edge cases, gradually build trust in AI systems, and provide feedback for model improvement.
Definition
Inference is the process of applying a trained AI model to new data to generate outputs, such as classifying a part as OK or defective or predicting a failure risk score.
Why it matters on the shop floor
Inference is what runs in real time next to the line or on a dock—turning sensor readings and images into decisions without retraining the model each time.
Definition
The Internet of Things (IoT) is the network of connected physical devices—such as sensors, actuators, and controllers—that collect and exchange data.
Why it matters on the shop floor
IoT devices provide the raw data for quality monitoring, predictive maintenance, energy optimization, and many other AI-enabled applications in manufacturing.
Definition
A label is the “answer” attached to each training example, such as “defective,” “OK,” or a numeric value like measured viscosity, that the model is trained to predict.
Why it matters on the shop floor
Accurate, consistent labeling of defects, failures, or quality measurements is essential for training reliable inspection and predictive models.
Definition
Latency is the time delay between an input (such as a sensor reading or camera trigger) and the system’s response or decision.
Why it matters on the shop floor
Low latency is essential for real-time applications such as ejecting defective parts, controlling processes, or verifying pallets as they move through a dock portal.
Definition
A large language model (LLM) is a type of AI model trained on very large amounts of text so it can understand and generate language, answer questions, and follow natural-language instructions.
Why it matters on the shop floor
LLMs can power assistants that explain alarms, suggest next troubleshooting steps, draft reports, or make technical information more accessible to engineers and operators using everyday language.
Definition
Machine learning (ML) is a subset of AI that uses data and algorithms to learn patterns and make predictions or decisions without being explicitly programmed for every rule.
Why it matters on the shop floor
ML underpins most modern quality, maintenance, and optimization solutions, turning historical and live data into models that can be deployed on lines and equipment.
Definition
Natural language processing (NLP) is a branch of AI that enables computers to understand, interpret, and generate human language in text or speech.
Why it matters on the shop floor
NLP powers assistants and tools that can search documentation, summarize reports, or answer questions about processes and alarms in everyday language.
Definition
A neural network is a computational model made up of layers of interconnected “neurons” that learn to transform inputs into useful outputs by adjusting internal weights.
Why it matters on the shop floor
Neural networks are the basis for many vision, predictive maintenance, and analytics models that can learn complex relationships in manufacturing data.
Definition
Noise is unwanted variation or randomness in data, such as sensor glitches, electrical interference, or visual artifacts in images.
Why it matters on the shop floor
Handling noise correctly—through filtering, robust models, and good sensor practices—is key to building AI systems that work reliably in real industrial environments.
Definition
Overfitting occurs when a model learns patterns that are too specific to the training data, performing well on that data but poorly on new, real-world data.
Why it matters on the shop floor
An overfitted inspection or maintenance model may look great in testing but fail when lighting changes, a new product variant is introduced, or conditions drift.
Definition
Precision measures, out of all the items the model flagged as defective (or positive), how many were actually defective—high precision means few false alarms.
Why it matters on the shop floor
In visual inspection and dock verification, good precision reduces nuisance rejects and rechecks, so operators spend less time dealing with false positives and more time on real issues.
Definition
Predictive maintenance uses historical and real-time data from machines (for example vibration, temperature, current, or pressure) to estimate equipment health and predict when failures are likely to occur.
Why it matters on the shop floor
Instead of waiting for breakdowns or relying only on fixed time-based schedules, predictive maintenance helps schedule work during planned stops, reduce unplanned downtime, and extend asset life.
Definition
Recall measures, out of all the items that were actually defective (or positive), how many the model correctly caught—high recall means few misses.
Also called sensitivity or true positive rate
Why it matters on the shop floor
High recall is critical when missing a defect is costly or unsafe; it shows whether the system is catching nearly all problem units before they move downstream or to the customer.
Definition
A regression model is an AI model that predicts a continuous numeric value, such as remaining useful life, viscosity, or temperature, rather than a discrete category.
Why it matters on the shop floor
Predictive maintenance and process optimization often rely on regression models to estimate things like failure risk or critical process parameters ahead of time.
Definition
Retraining means updating an existing model with new data to improve performance or adapt to changes in products, processes, or conditions.
Why it matters on the shop floor
Planned retraining helps AI systems stay accurate over time as new variants are introduced, equipment ages, or upstream processes change.
Definition
SHAP (SHapley Additive exPlanations) is a method for explaining model predictions by assigning each input feature (for example, vibration level, temperature, or line speed) a numerical value that shows how much it pushed a particular prediction up or down compared to a baseline.
Why it matters on the shop floor
SHAP helps engineers and domain experts see why an AI model flagged a product as defective or predicted a higher failure risk—making it easier to trust the system, debug unexpected behavior, and link model decisions back to real process conditions.
Definition
Specificity measures, out of all the items that were actually good (or negative), how many the model correctly identified as good—high specificity means few good items are flagged as bad.
Why it matters on the shop floor
Specificity helps quantify how often good products are being incorrectly rejected, which ties directly to scrap, rework, and line efficiency.
Definition
Supervised learning is a type of machine learning where a model is trained on labeled data—inputs paired with known outputs—so it can learn the relationship between them and make predictions on new data.
Why it matters on the shop floor
Most inspection and predictive maintenance models are trained using supervised learning, based on past examples of good and bad parts or normal and faulty behavior.
Definition
Synthetic data is artificially generated data—such as simulated images, defects, or sensor signals—created to supplement or replace real-world data for training and testing models.
Why it matters on the shop floor
Synthetic defect images or simulated failure signals can help train models when real examples are rare, hard to capture, or risky to reproduce in production.
Definition
Time series data is a sequence of measurements taken over time at regular or irregular intervals, such as vibration, current, temperature, or pressure readings from a machine.
Why it matters on the shop floor
Predictive maintenance and process monitoring rely heavily on time series data to detect trends, cycles, and early warning signs of failures.
Definition
A training set is the portion of labeled data used to teach a model, containing examples where the correct outcome (label) is already known.
Why it matters on the shop floor
The quality and representativeness of the training set strongly influence how well an AI system will perform on real production data.
Definition
Transfer learning is a technique where a model trained on one task or dataset is adapted to a related task, reusing much of what it has already learned.
Why it matters on the shop floor
Transfer learning can reduce the amount of data and time needed to bring AI solutions to new lines, products, or plants by building on existing models.
Definition
In manufacturing quality control, a true negative occurs when a product that is actually good is correctly classified as good.
Why it matters on the shop floor
True negatives represent the normal case for high-yield processes; they help quantify how well the system avoids unnecessary rejects.
Definition
In manufacturing quality control, a true positive occurs when a product that is actually defective is correctly classified as defective.
Why it matters on the shop floor
True positives show how many real defects the system successfully catches before they move downstream or to customers.
Definition
Underfitting occurs when a model is too simple to capture the real patterns in the data, leading to poor performance both on training data and in production.
Why it matters on the shop floor
An underfitted model might miss obvious defects or equipment issues, giving a false sense of security about the performance of an AI solution.
Definition
Unsupervised learning is a type of machine learning that finds patterns or structure in unlabeled data, without predefined output labels.
Why it matters on the shop floor
Unsupervised methods can help discover natural groupings in sensor or process data, detect anomalies, and suggest where to focus deeper investigation even when labels are scarce.
Definition
A validation set is a portion of labeled data that is held out from training and used to tune model settings (for example, thresholds or hyperparameters) and check how well a model generalizes before it is tested on completely unseen data.
Why it matters on the shop floor
A good validation set—representative of real production conditions, including typical variation and known defects—helps ensure that an AI model for visual inspection, dock checks, or predictive maintenance performs reliably in practice, not just on the data it was trained on.
Definition
Visual inspection in an AI context means using cameras and computer vision models to examine products or surfaces for defects, instead of (or in addition to) manual human inspection.
Why it matters on the shop floor
Automated visual inspection can look at every unit, apply consistent criteria across shifts and plants, and detect defects that are hard for tired human eyes to see at high speed.