DaoAI SkyVision 0-code video surveillance AI platform (on-site hourly training of proprietary models, behavior/event recognition, edge box real-time alerting, 100% on-premise data security, DaoAI World Model semantic understanding) precisely identifies critical operation omissions in assembly processes, reducing the assembly missed detection rate for a leading automotive parts supplier from an average of 2.5% to below 0.8%, significantly improving product quality and production efficiency.
Industrial SOP (Standard Operating Procedure) / operational guidelines are the cornerstone of quality control in modern manufacturing. In increasingly complex industrial assembly processes, ensuring that every step is strictly followed is critical for product quality and safety. Especially in industries like precision machinery or automotive component manufacturing, a subtle assembly step omission can lead to product functional failure, recall risks, or even safety incidents. A leading automotive parts supplier, while producing a new electronic control unit (ECU), had an assembly line with over 30 critical steps. A particular key connector installation process, due to confined operating space, severe visual obstruction, and fast manual operation pace, was highly prone to omissions or incomplete fastening. Traditional quality inspection methods were often lagging, making it difficult to detect problems in real-time and correct them promptly, resulting in batches of defective products flowing to subsequent processes or even the market, causing significant losses.
Pain Points: Why This Hurdle Is So Difficult to Overcome
This client faced multiple challenges in the critical connector assembly process. Firstly, a missed detection rate as high as 2.5% meant that for every 1000 ECUs produced, 25 might have issues with missing or incomplete connector fastening, severely impacting product reliability. Secondly, traditional manual visual inspection or sampling methods were inefficient, couldn't cover all products, and were susceptible to operator fatigue, leading to fluctuating false alarm rates, with an average of over 100 hours of manual re-inspection required monthly due to false alarms. Thirdly, the assembly cycle time for this process was extremely demanding, with each product staying for only 15 seconds, leaving a very short window for inspection. Furthermore, the connector installation position was deeply embedded within the product, surrounded by numerous reflective metal components, posing significant challenges to the imaging stability and recognition accuracy of traditional vision solutions. The current industry hot topic of how multimodal intelligent cameras can enhance the proactivity and intelligence of smart security systems through “dialogue” and “doing work” capabilities also inspired us to shift from passive monitoring to active identification and early warning. However, how to apply this “dialogue” and “doing work” capability to millisecond-level production line behavior recognition was a pressing problem.
The root causes of these difficulties included: complex processes leading to high operational variability, which increased recognition difficulty; under high cycle times, traditional visual rules struggled to adapt to lighting changes and subtle posture differences; manual visual inspection, when operators were fatigued, often missed subtle omissions, and data could not be closed-loop for continuous optimization. These factors collectively led to persistently high missed detection rates, becoming a key bottleneck restricting the improvement of production line efficiency and product quality.
Technical Principles
DaoAI SkyVision 0-code video surveillance AI platform, through its core DaoAI World Model semantic understanding capability, achieves deep understanding and precise recognition of industrial assembly behaviors. The platform uses multimodal intelligent cameras as front-end data collection devices, with built-in edge computing capabilities for preliminary data processing and AI model inference. We have moved away from traditional rule-based or feature engineering visual detection methods, instead leveraging deep learning and large model technologies. By learning from a large amount of video data of normal and abnormal assembly behaviors, we construct an AI model capable of recognizing whether a “connector has been installed in place.” During on-site deployment, the SkyVision platform supports hourly training of proprietary models, meaning clients only need to provide a small amount of on-site data to quickly iterate and optimize the model, making it precisely adapt to the complex environment of specific processes, reducing the missed detection rate to <0.8%.
Compared to traditional methods, SkyVision's advantage lies in its adaptability and generalization capabilities. Traditional rule-based AOI solutions require engineers to manually set complex geometric rules and thresholds, which are sensitive to changes in lighting, product posture, material types, etc. Once the production line changes models or the environment shifts, it requires significant time for reprogramming and debugging, and cannot effectively handle subtle behavioral judgments like the degree of connector fastening. In contrast, DaoAI SkyVision's deep learning-based model can extract high-level semantic features from multi-dimensional video streams, such as the operator's trajectory, the relative position of the connector to the product body, and the deformation characteristics of the connector at the moment of fastening, thereby achieving precise differentiation between “installed” and “not fully installed.” The DaoAI World Model further enhances this semantic understanding, enabling it to not only recognize individual events but also understand the context in which they occur, effectively filtering out false alarms. This reduces the false alarm rate by over −70%, significantly outperforming traditional methods and saving clients a considerable amount of manual re-inspection time.
Typical Application Scenarios
- **Bolt/Nut Omission Detection:** In heavy machinery assembly like engines or transmissions, missing bolts or nuts can lead to serious accidents. SkyVision ensures every critical fastener is installed by identifying the action sequence of tightening tools and the presence of the bolt itself. The challenge lies in the small size and large number of bolts, and potential obstructions during tightening.
- **Wiring Harness Insertion Omission Detection:** In industries such as automotive electronics and home appliances, correct wiring harness connections are crucial. The platform can monitor the operator's action of inserting a wiring harness plug into a designated socket and determine if it is fully engaged. Challenges include diverse plug colors, hidden socket locations, and the flexibility of harnesses.
- **Seal Ring/Gasket Omission Detection:** In the assembly of waterproof, dustproof, or leak-proof products, missing seal rings or gaskets can cause product function failure. SkyVision can identify the operator's actions of picking and installing seal rings, and determine if the seal ring is in the correct position. Difficulties include seal rings often being black, thin, easily deformable, and having low contrast with the background.
- **Clip/Retainer Omission Detection:** Clips and retainers commonly used in plastic or metal component assembly, if omitted, can affect the fixing strength of parts. The platform ensures complete installation by identifying the insertion action and final state of the clip. Challenges include small clip size and potential concealment by other components after installation, requiring monitoring from multiple angles.
- **Label/Identification Sticker Omission Detection:** On product packaging or critical components, missing labels can affect product traceability and compliance. SkyVision can monitor the entire process of label printing, removal, and application, and determine if the label has been affixed to the designated position. Difficulties include diverse label materials, reflections, and potentially irregular application positions.
Case Study
Before adopting the DaoAI SkyVision platform, a leading automotive parts supplier had long been plagued by connector installation omissions on its new ECU production line. Traditional manual sampling and end-of-line functional tests could not effectively prevent defective products from being released, leading to an average of 2.5% of products requiring rework or scrap each month, with customer complaints occurring frequently. To address this pain point, they decided to deploy the DaoAI SkyVision 0-code video surveillance AI platform. After project initiation, the DaoAI team first evaluated the on-site workstations, installed multimodal intelligent cameras, and collected a small amount of video data of normal and abnormal assembly behaviors. Leveraging SkyVision's hourly model training capability, engineers completed the AI model training and initial deployment for this specific connector installation process in less than a day. After two weeks of online trial operation and data tuning, the platform successfully increased the detection rate of connector omissions to over 99.2%, stably reducing the missed detection rate to below 0.8%. Furthermore, through the edge box real-time alerting mechanism, once an omission was detected, the system immediately triggered an audible and visual alarm and stopped the production line, effectively preventing defective products from flowing to the next process, reducing the rework rate due to omissions by −68%. This significantly improved production line efficiency and product quality. The client stated that DaoAI SkyVision's 100% on-premise deployment ensured data never left the facility, fully complying with their strict data security requirements.
“DaoAI SkyVision not only solved our long-standing assembly omission pain point, but more importantly, it showed us the immense potential of AI vision in understanding production line behaviors, truly achieving a leap from ‘seeing’ to ‘understanding’.” — Production Manager, Automotive Parts Supplier
DaoAI Solutions and Products
The core solution provided by DaoAI to this client was an assembly step omission detection system based on the SkyVision 0-code video surveillance AI platform. The platform's key advantage lies in its “0-code” characteristic, allowing production line engineers, without professional AI development background, to train and deploy AI models on-site using a simple graphical interface and a small amount of sample data. For connector installation omissions, we utilized SkyVision's behavior recognition module, deploying lightweight models on the edge box side to achieve millisecond-level real-time inference. When an operator fails to complete the connector installation according to SOP, the edge box immediately sends an alarm signal to the production line control system via an integrated interface, triggers an audible and visual alarm, and locally saves abnormal video clips for subsequent traceability and analysis. The entire system supports 100% on-premise private deployment, with all video data and model inference results processed within the client's factory premises, ensuring data security and privacy compliance. Additionally, the DaoAI World Model, as a unified foundation, endows SkyVision with stronger semantic understanding and cross-scenario generalization capabilities, enabling it to continuously learn from production line feedback, constantly improving model accuracy and robustness, ensuring long-term stable operation of the system.
Through the DaoAI SkyVision platform, the client achieved refined management and real-time monitoring of assembly processes. Not only was the missed detection rate for critical processes reduced to <0.8%, but the rework rate due to omissions was also reduced by −68%, effectively saving rework costs and time. Concurrently, the significant reduction in false alarm rate (reduced by −70%) freed up substantial manual re-inspection resources, allowing production personnel to focus on higher-value tasks. This proactive, intelligent quality control method significantly improved product first-pass yield and customer satisfaction, bringing tangible business value to the enterprise.
FAQ
How does SkyVision achieve on-site hourly model training?
DaoAI SkyVision platform features built-in APDT few-shot self-training technology and a 0-code model training interface. Users don't need programming; they simply upload a small number of on-site collected video clips of normal and abnormal behaviors via a graphical interface. The platform then leverages the pre-trained DaoAI World Model for rapid transfer learning and fine-tuning. Typically, a dedicated AI model for a specific process can be generated and deployed within hours, significantly reducing deployment cycles and technical barriers.
How does the platform handle complex visual obstructions and lighting variations in industrial environments?
SkyVision platform uses multimodal intelligent cameras for data acquisition, combined with the powerful semantic understanding capabilities of the DaoAI World Model. It can analyze video streams from multiple angles and dimensions. Even with partial obstructions or uneven lighting, the model can accurately determine assembly status by recognizing advanced features such as operator auxiliary movements and subtle component deformations. The model also possesses a certain degree of robustness, adapting to minor fluctuations in the on-site environment.
How are data security and privacy ensured with SkyVision?
DaoAI SkyVision platform supports 100% on-premise private deployment. All video data, model training data, and inference results are processed and stored on edge boxes or local servers within the client's factory premises. Data is not uploaded to any external cloud platforms, fundamentally eliminating the risk of data leakage. This ensures strict protection of the client's sensitive production data and intellectual property, fully complying with the highest requirements for data security and privacy in the industrial sector.