Many factories, when mentioning AI visual recognition for shipbuilding, immediately think of simply mounting an algorithm box at the workstation and considering the task done. The problem lies here: while the recognition model is installed, the work instructions at the workstation remain on paper. Workers look at paper while AI looks at the camera; the two information streams operate independently. Video SOP is not a side dish; it is the foundation for whether AI visual recognition can be successfully implemented at the workstation.
1. How Video SOP is Used at the Workstation
The shipbuilding scenario is unique: segments are large, processes are long, and assembly relationships are complex. When paper SOPs are posted on walls, workers must run back and forth to check them. When PDF work instructions are viewed on touchscreens, oily hands make it impossible to swipe through pages. Video SOP solves this by recording key actions such as assembly sequences, welding postures, and coating thickness control into short videos that play on loop or on demand directly on the all-in-one workstation terminal.
Workers do not need to search for files or log into systems; the video SOP corresponding to the current work order appears immediately upon startup. The camera for shipbuilding AI visual recognition is installed at the same workstation, and the recognition results are displayed on the same screen as the video SOP. If AI determines the assembly is complete, the corresponding frame in the video is automatically tagged. If AI triggers an alarm, the worker can look up to see the standard action in the video for comparison.
2. Video SOP Production: What to Film, How to Film, and Who Manages It
Filming a video is not about creating a promotional film. The video materials required for shipbuilding AI visual recognition have three hard requirements. First, the camera angle must be fixed and align as closely as possible with the AI camera's perspective so workers are not confused when comparing. Second, each video should be controlled between 30 to 60 seconds, covering only one action or one inspection point, such as a separate video for "rib assembly tack welding sequence" and another for "weld seam appearance inspection standards." Third, the video must include watermarks showing the workstation number, process number, and version number, which AI uses to match and retrieve the correct content.
In terms of the production workflow, after the process engineer films the video using a mobile phone or industrial camera, they upload it directly to the ESOP management end and attach it to the corresponding process node. The system supports split-screen playback: the left screen plays the video SOP, while the right screen displays the real-time AI recognition image and judgment results. Suppose a hypothetical ship segment assembly line where a worker is at the rib positioning station; the left screen loops the positioning weld sequence video, while the right screen uses AI visual recognition to draw boxes around weld points in real time and compare them against standard coordinates. Any deviation is immediately highlighted in red. This linkage requires no additional interface development; the native ESOP all-in-one terminal supports it.
3. Playback and Split-Screen: Practical Points at the Workstation
There are two pitfalls to avoid when playing video SOPs at the workstation. First, automatic playback must not interrupt the worker's operational rhythm; videos default to silent looping, and workers can tap a speaker icon if sound is needed. Second, split-screen layouts must be preset according to process types. Assembly stations use a left-right split screen with "video + AI recognition image," while inspection stations use a top-bottom split screen with "video + inspection data Kanban."
Judgment results from shipbuilding AI visual recognition link with the Andon module in ESOP. If recognition indicates non-compliance, the station Andon light turns yellow, and the video SOP automatically jumps to the standard disposal video segment for that specific defect. Simply put, AI is responsible for detecting anomalies, while video SOP is responsible for instructing the worker on how to correct them; both actions are completed on the same screen.
4. Training Video Management: Do Not Let Videos Sit in Folders
Producing the video SOP is not the end goal. In the shipbuilding industry, personnel turnover and cross-skilling are frequent, so training videos must be bound to specific positions. The ESOP training module supports pushing video courses based on workstation, process, or personnel role. After watching the video, workers confirm completion online, and the record is saved to their personal training file. The management end can track who has not watched or confirmed. Before a production line change or the start of a new ship type, videos can be pushed in batches. Scenario estimation suggests this can compress the training preparation cycle by approximately half, though actual results depend on the completeness of the factory's basic data.
Video version management is equally critical. After the model for shipbuilding AI visual recognition iterates, the corresponding standard action videos must be updated simultaneously. ESOP document signing and version control cover video files; old versions are automatically archived, and workstation terminals are forced to refresh to the latest version. This prevents a mismatch where AI uses a new model while workers view old videos.
5. Implementation Suggestions
Start by selecting a typical process for a pilot: choose a workstation where AI visual recognition is already running but workers still rely on paper SOPs for reference. Film three core action videos and configure a left-right split-screen template. Run this for two weeks to gather worker feedback.