#01 What are visual inspection devices?
Visual inspection refers to the process of using machine vision products (i.e., image acquisition devices, which are divided into CMOS and CCD types) to convert the captured target into an image signal, which is then transmitted to a dedicated image processing system. Based on pixel distribution and information such as brightness and color, the signal is converted into a digital signal. The image system performs various operations on these signals to extract the features of the target, and then controls the operation of the equipment on site based on the judgment results.
With the rapid implementation of artificial intelligence technology and the continuous development of the intelligent robot industry, visual inspection machines are showing even stronger vitality.
The typical structure of a visual inspection device consists of five main parts: illumination, lens, camera, image acquisition, and motion control system.
#02 What testing phases do visual inspection projects go through?
1. Software Testing
Focus on whether the implementation of functions is consistent with customer needs, ensure the correctness of the software process, and the correct application logic.
By controlling the input to the software, different test results can be achieved. By comparing the differences between the input and output, the correctness and accuracy of the test can be verified, thereby discovering vulnerabilities in the system and initiating a cycle of research and development modification and testing verification.
2. Hardware Testing
Conduct hardware reliability testing on the hardware itself and the environment, such as aging testing, compatibility testing, and failure rate testing.
It is necessary to test common hardware configurations to determine whether the software can run in a variety of hardware environments, such as compatibility testing with the main unit, boards, and peripherals.
3. Joint debugging and testing
The main focus is on testing the hardware and software integration functions: verifying the correctness of electrical and software signal communication logic, light sources, cameras, and other hardware-triggered functions such as taking pictures and scanning, and statistically analyzing the detection results.
When interfacing with external systems, it is necessary to check whether the basic functions of the modules, algorithm detection capabilities, and cycle time of each system meet the customer's expected standards.
4. Model Testing
For model testing, the focus is on functional testing, performance testing, evaluation of model metrics, and analysis of metric results.
Methods employed: A/B testing, robustness testing, offline testing, and online testing were used to evaluate the reliability and stability of the model.
Offline/Online Testing: Historical data is used for offline training and testing evaluation to generate an offline model version. To ensure the stability of the model service, online and offline consistency verification and online stability verification testing are required.
Consistency test: Using the same batch of samples, the system requests the model service in both offline and online environments, and then compares and analyzes the decision results of the model service.
Stability testing: The model is evaluated based on the decision results obtained from online requests using new samples. Model evaluation metrics include KS and ROC.
A/B testing: To ensure stability, the algorithm model will iterate and generate multiple versions. If the online service uses model A for decision-making, when it is necessary to iterate the model version, model B needs to be trained in an offline environment, then deployed to the online environment, and the results of online and offline A/B tests are analyzed. The conversion rate and other methods are used to help decide which version to use.
5. Performance Testing
The software simulates different user volumes to conduct interface performance tests.
The hardware continuously acquires images to verify the stability of the program, the stability of signal communication, and the correctness of the image storage logic.
Hardware and software integration stability testing: Based on the customer's production environment deployment scenario, implement load operation for a certain period of time, monitor and analyze whether the cycle time and performance indicators meet the customer's expectations.
#03 How to conduct testing of visual inspection equipment?
1. Product Requirements Analysis
Requirements analysis is the first step in starting testing. Based on the project's requirements, testers need to extract the project's testing scope, functionalities, business processes, expected outputs, exception handling, and the required time and resources.
2. Test point disassembly
A test case is a set of data consisting of preconditions, test inputs, execution conditions, and expected results, used to perform tests on a specific requirement or objective. It is a document that reflects the test plan, methods, techniques, and strategies.
Test engineers need to consider the specific requirements of a project from mechanical, electrical, software, and algorithm perspectives to design test cases.
3. Test Execution and Defect Management
Differentiate the testing phases: Execute test cases separately and use testing tools to assist in testing and improve testing efficiency, such as camera debugging tools, PLC debugging software, database tools, interface and performance testing tools, etc.
Definition of defect:
(1) The functional requirements stipulated in the contract or technical agreement have not been met;
(2) An error occurred that was not specified in the product manual;
(3) The functions exceed the scope of the product manual;
(4) The product failed to meet the objectives that should be achieved, even though not specified in the product manual;
(5) The defect is considered difficult to understand, difficult to operate, slow to run, or poorly received by end users. Test engineers need to clearly define the type, priority, and severity of defects, and use defect management tools or defect lists to record, provide feedback on, and verify defects.
4. Verification of key inspection items
The following points should be considered when verifying goods before shipment:
Safety check of the line: The ground wire must be installed first.
When visual inspection equipment is in operation, high voltage is present. To avoid safety accidents, installing a grounding wire during periods of frequent use of the visual inspection device will significantly improve safety measures.
Voltage stability check: The operating voltage of general industrial equipment is 220V, and some are 380V.
Voltage imbalance can easily cause serious damage to industrial equipment and reduce the service life of machinery.
Interface checks: such as digital inputs and outputs, starters used for triggering, communication with PLCs, robots, linear axes, database systems, etc.
Debugging and verification: System testing using a small number of workpiece variables for both manual and automatic operation.
Results Record: Good and bad images should be recorded in detail, along with an evaluation of the test results.
#04 Requirements for a High-Quality Machine Vision Inspection System
1. High-quality imaging system
Imaging systems are often referred to as the "eyes" of visual inspection equipment; therefore, the quality of the "eyes'" recognition ability is the most critical indicator for evaluating imaging systems. Typically, the evaluation metrics for imaging systems are mainly reflected in three aspects:
(1) Whether the existing defects can be discovered
The most basic and only data that can be relied upon for detection based on image methods is the color (or brightness change) on the captured image. There is no other data available for reference.
Therefore, a high-quality imaging system should first and foremost be one that can fully represent the color changes on the surface of the object being inspected.
Therefore, in addition to choosing high-definition cameras and lenses, the lighting design used to create the imaging environment is also very important. Sometimes, lighting systems are even designed specifically for special defects.
The commonly referred to 100% quality inspection system actually refers to 100% full inspection of images that can fully represent various defects.
(2) The smallest size of the defect that can be detected
The smallest unit of measurement for a digital image is the pixel, which does not represent the actual size of the subject.
The relationship between the actual size of the subject and the number of pixels is established through a physical quantity called resolution. Resolution refers to the actual size of the object represented by each pixel.
The smaller the resolution value, the higher the image detail, the smaller the defect size that the detection system can detect, and the higher the detection accuracy.
(3) Can the image be captured quickly enough?
Just like how the human eye perceives moving objects, when an object moves fast enough, the human eye can no longer clearly observe the entire object.
The camera, the "eye" of a machine vision inspection system, also has a shooting speed limit, namely the camera's main frequency.
When the speed of the object being photographed exceeds the camera's maximum clock speed, the camera cannot obtain a clear and complete image, and the detection cannot continue normally.
The higher the camera's clock speed, the faster the data acquisition speed, ensuring efficient detection.
Therefore, whether a camera with a sufficiently high clock speed is used is also a key factor in evaluating the quality of an imaging system.
2. Mature image processing and analysis algorithms
Image processing and analysis algorithms in the entire detection system are equivalent to the human brain's judgment and thinking during manual detection.
Since machine vision is a highly practical discipline, evaluating the quality of an algorithm relies more on verification through real-world applications than on whether the algorithm employs advanced or complex theories.
Therefore, what we need is a stable and efficient image processing and analysis algorithm that can fully simulate the human brain's judgment process and methods, which is what we call a mature processing and analysis algorithm.
Therefore, when designing processing algorithms, it is necessary to fully analyze the human judgment process and translate it into the language of the computer.
3. Good operability
Good operability primarily requires that the application and operation of the testing equipment be simple, convenient, and easy to understand. For example, the system should have a user-friendly human-machine interface and a well-designed directional operating system.
4. Stable supporting facilities
Other supporting facilities refer to facilities other than the detection system, such as the transmission control platform and defect handling devices (rejection, alarm, marking, etc.). They must operate stably and respond to signals promptly and quickly.
