The Chinese edge computing platform can complete image collection, inspection area management, assembly difference determination, result recording and equipment interface interaction at the production site. Whether the system can be used in actual workstations depends not only on whether the algorithm can recognize it, but also on whether imaging conditions, workpiece positioning, trigger timing, exception handling and acceptance data form a closed loop.
Winge Technology has completed the actual machine adaptation verification of a set of assembly visual inspection software and algorithm applications on Huawei Ascend Atlas 200I DK A2, supporting USB/UVC cameras, reference image, multiple ROI rules, Chinese and English interfaces, result evidence and version rollback. Subsequent complete machines will be equipped with Chinese cameras, lenses, light sources, I/O and a software environment that can be used for formal projects in a fully China-localized direction.
What is Chinese edge computing assembly visual inspection?
This type of system deploys image collection, algorithm judgment and workstation interaction on a Chinese computing platform at the production site. The goal is to determine whether a part is present, has the correct model or orientation, and whether there are identifiable offsets or anomalies in critical areas for a predefined assembly location.
"Chinese edge computing platform" describes the computing and deployment form; "Assembly integrity and incorrect and missing assembly detection" describes business tasks. The project still needs to select cameras, lenses, light sources, algorithms and sensors based on the detection objects. The feasibility cannot be judged based on the name of the computing platform alone.
Why deploy detection on the workstation side?
Deployment on the workstation side can complete mapping and judgment locally, which facilitates the formation of a definite interactive relationship with trigger signals, sound and light prompts, PLC or other control devices. Whether images are uploaded to the server, how long they are retained, and who accesses them should be determined by the project's data policy.
Edge deployments do not automatically equate to low latency, high security, or offline availability. The project must actually measure single detection time, concurrent tasks, storage usage, network disconnection behavior and permission configuration to form corresponding conclusions.
What parts does the system consist of?
A typical Chinese assembly visual inspection system includes the following parts:
- Chinese edge computing equipment: running collection, algorithm, interface and interface services;
- Chinese industrial cameras: obtain workpiece images according to trigger conditions;
- Chinese lenses and light sources: form repeatable and distinguishable target features;
- Detection software: manages reference image, ROI, thresholds, tasks and user operations;
- Algorithm module: perform reference graph comparison, rule determination or model reasoning;
- Chinese I/O, PLC or controller: transfer in place, trigger, PASS, FAIL and abnormal status;
- Result storage: associated time, task, artifact identification, judgment and evidence image;
- Operation and maintenance mechanism: supports logs, configuration backup, version upgrade and rollback.
The website does not need to disclose the complete BOM, but the formal implementation must use actual equipment and software versions to complete joint debugging and acceptance.
How does assembly error detection work?
1. Establish stable imaging conditions
First determine the field of view, working distance, target pixels, exposure and light source angle. Camera parameters should be controlled during production runs to avoid automatic exposure or changes in ambient light causing different images of the same workpiece.
2. Establish reference image and detection area
Use the qualified workpiece to establish a reference image, and then configure ROI for screws, connectors, labels, wire harnesses, buckles or other key locations. Each ROI should correspond to clear detection objects and judgment rules.
3. Execute algorithm judgment
When fixed stations and position changes are small, reference drawings, templates, colors, edges or geometric rules can be used. Classification, detection, or segmentation models can be evaluated when there are large changes in categories, poses, or backgrounds.
4. Output interpretable results
The result should not be just one PASS or FAIL. The system should also save the failure area, rule name, score or threshold, original picture or evidence picture, and necessary exception reasons to facilitate review and traceability.
5. Linked with workstation process
The test results are comprehensively processed by PLC, safety control system or on-site process. Vision software can provide decision signals, but actions involving equipment shutdown, interception, or personnel safety must be handled by risk-assessed control logic.
How to choose between traditional vision and learning algorithms?
Traditional vision is suitable for tasks with clear features, stable positions, and a limited number of samples. Development and interpretation costs are usually more controllable. Learning models are suitable for tasks with many categories, large appearance changes, or tasks that require learning complex features, but require training data, model deployment, interface access, and consistency verification.
Currently, Winge Technology assembly inspection software and algorithm applications have been run on Huawei Ascend Atlas 200I DK A2, and verification of cameras, standard images, ROI rules, inspection results and version rollback links have been completed. For specific projects, reference image comparison, rule algorithms or learning models can be selected based on the sample and detection target.
What is the role of Atlas 200I DK A2 in solution verification?
Atlas 200I DK A2 is the Ascend AI development kit for developers. According to official product specifications, its AI processor supports 8 TOPS INT8 or 4 TFLOPS FP16 computing power and provides 4GB of memory. This parameter is used to describe the platform specifications and does not mean that this project has reached the corresponding application performance.
The developer kit is suitable for verifying the application structure, camera access, interface, algorithm interface and deployment method. Formal commercial projects also need to confirm commercial hardware, drivers, firmware and operating environment based on the official software usage statement and target devices.
What needs to be confirmed for a pure localization plan?
Pure localization implementation does not require the disclosure of BOM, but the following confirmation needs to be completed within the project:
- The actual configuration of the computing platform, camera, lens, light source and I/O;
- Sources and licenses of operating systems, drivers, runtimes, and third-party dependencies;
- Camera interface, triggering method, field protocol and workstation timing;
- Sample range, test objects and allowed appearance changes;
- How to handle exceptions, network outages, power outages and equipment disconnections;
- Version upgrade, data backup, log retention and rollback methods.
Before the joint debugging and acceptance of the complete system is completed, it is recommended to use "implemented according to a fully China-localized plan" or "Chinese configuration will be used" instead of "fully China-localized system has been completed" in advance.
Which scenarios are suitable for this type of solution?
- Inspection of connectors, screws, snaps and labels on electronic components;
- Part presence, orientation and position inspection of mechanical assemblies;
- Inspection of wire harness insertion, color, direction and fixation status;
- Complete set of accessories and inspection of appearance condition before packaging;
- Assembly stations where evidence of inspection needs to be recorded locally.
Objects that are transparent, strongly reflective, flexible, move at high speed, or are severely obscured require imaging and sample verification first, and detection results cannot be guaranteed based on the scene name alone.
What does the client need to provide before the project starts?
Customers need to provide at least:
- Products inspected, assembly processes and error types;
- Qualified samples, defective samples and boundary samples;
- Workstation space, rhythm, field of view and ambient light conditions;
- Existing PLC, I/O, MES or other interface requirements;
- Whether it is necessary to save the original image, retention period and access rights;
- Acceptance indicators, statistical caliber and retest rules.
If there is a lack of real defect samples, small-scale on-site verification can be designed first, but the reference image demonstration results cannot be used to replace the final acceptance.
How to accept the Chinese edge assembly visual inspection system?
For acceptance, a sample set independent of the development and parameter adjustment process should be used to separately count false positives of qualified products, missed detection of defective products, inability to determine, and system abnormalities. It is also necessary to record the continuous running time, single detection time, trigger consistency, evidence integrity, disconnection recovery and version rollback results.
Accuracy, missed detection rates, and beats must also account for the number of samples, defect types, collection conditions, and statistical methods. When there is no test data, promotional conclusions such as "high accuracy", "zero missed detections" or "real-time detection" should not be used.
FAQ
Do Chinese edge computing platforms have to use AI models?
Not required. Tasks with fixed positions and clear differences can use traditional vision; complex classification, detection or segmentation tasks re-evaluate the learning model. The algorithm route should be determined by samples and acceptance metrics.
Does a purely localized plan have to disclose the complete BOM?
No need. The website can explain the scope of localization and implementation methods, and specific brands, models and supply information can be confirmed during the project stage. The joint debugging of the complete system still needs to be completed based on the real configuration.
Can visual results directly control equipment shutdown?
Visual results can serve as an input to the control system. Critical shutdowns and safety actions should be handled by the PLC or safety control system in conjunction with other conditions, isolation circuits, failure modes and on-site risk assessment.
Is the current detection software already running on the Huawei Ascend Atlas platform?
Already running. The current version has completed software and algorithm application deployment on Atlas 200I DK A2, as well as camera access, ROI rules, result recording and version rollback verification.
Can the detection accuracy be directly given?
It cannot be given without samples and working conditions. The accuracy rate needs to be calculated based on the independent test collection confirmed by the customer, and the handling methods of false positives, missed detections, undetermined and abnormal samples should also be explained.
Official information and factual sources
- Huawei Ascend: "Atlas 200I DK A2 Developer Kit Product Introduction"
https://www.hiascend.com/document/detail/zh/Atlas200IDKA2DeveloperKit/23.0.RC2/pd/pd_0001.html
- Huawei Ascend: "Atlas 200I DK A2 Basic Specifications"
https://www.hiascend.com/document/detail/zh/Atlas200IDKA2DeveloperKit/23.0.RC2/pd/pd_0004.html
- Huawei Ascend: "Atlas 200I DK A2 Software Usage Statement"
https://www.hiascend.com/document/detail/zh/Atlas200IDKA2DeveloperKit/23.0.RC2/qs/qs_0018.html
Related Atlas vision inspection resources
- Fully China-localized assembly vision inspection: Huawei Ascend Atlas project case study
- Winge Technology has completed on-device adaptation of assembly vision inspection on Atlas 200I DK A2
- What engineering work is included in Atlas 200I DK A2 vision inspection adaptation?
- From Atlas developer-kit validation to a China-localized production vision system
- How to deploy and deliver Huawei Ascend Atlas assembly vision inspection locally
- Huawei Ascend Atlas Vision Inspection
Online
Phone
WeChat
Top