What data looks like for this category
Data related to PCB manufacturing equipment inspection and maintenance comes from sources including Gerber files, drill files, and BOMs provided by customers, as well as factory-internal CAM files, equipment operating parameters, bath chemistry test records, AOI inspection reports, electrical test reports, and cross-section analysis reports. Update frequency follows per production batch, per shift, or continuous collection. Document structures vary significantly: engineering production instructions include process requirements and parameter thresholds; drilling and lamination parameter records include values such as rotational speed, pressure, and temperature; AOI and electrical test reports include fields such as defect location and size, with units including revolutions per minute, megapascals, degrees Celsius, millimeters, and others.
What constraints these characteristics impose on knowledge base retrieval and recall
Multi-source data isolated by production batch requires retrieval to bind to the corresponding batch’s dataset, to avoid irrelevant cross-batch data interfering with results. Diverse document types and specialized fields require the retrieval system to support multi-format parsing and specialized term matching. Frequently updated data requires the knowledge base to refresh regularly, to ensure recalled parameters match the current production batch. The accuracy of specialized parameters directly affects production quality, so the relevance and completeness of retrieval results must be strictly controlled to prevent missing key inspection and maintenance steps.
Configuration Settings
| Configuration Item | Suggested Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Specialized parameter and terminology matching for PCB manufacturing requires a high threshold, to avoid irrelevant generalized documents being included in recall results |
limit | 8–12 entries | Single-recall results need to cover multiple document types such as drilling, lamination, and electrical test. Too many results increase context processing load, while too few will miss critical information |
usingReRank | Enabled | FastGPT v4.17.0 supports this configuration. Reranking logic optimizes the ranking of specialized terms, returning more relevant maintenance steps and parameters first |
datasetSearchUsingExtensionQuery | Enabled | Expands abbreviated terms in user questions (such as AOI, Gerber) to cover more relevant historical documents and parameter records |
collectionIds | Bind the dataset ID corresponding to the production batch | Isolate data by production batch, to prevent irrelevant equipment parameters and inspection reports from cross-batch recall |
Three Common Mistakes
- Phenomenon: Retrieval result ranking does not match expectations, with results for specialized parameters appearing after irrelevant content. Cause:
usingReRankis not enabled, or thesimilaritythreshold is set too low, causing low-relevance generalized documents to be returned first. - Phenomenon: A
getaddrinfo EAI_AGAIN mongoerror occurs during knowledge base import, and the issue persists after restarting the service. Cause: The MongoDB connection configuration bound to the dataset was not updated with the production batch, or importing a large volume of production batch data triggered a connection timeout. - Phenomenon: Imported AOI report PDF images cannot be retrieved as text content. Cause: The image OCR function for document parsing is not enabled, or a parsing plugin that supports image formats is not configured, so detection data in images cannot be extracted as searchable text.
How to Confirm Configuration is Complete
- Submit a test query that includes specialized terms, check if recall results include documents from the corresponding production batch, and verify that the
collectionIdsbinding is effective. - Adjust the
similaritythreshold, observe changes in the relevance of recall results, and confirm that the threshold setting meets the matching requirements of the current business. - Import a PDF of an AOI report with images, verify that detection data in the images can be retrieved, and confirm that the image OCR function is enabled correctly.
- Import multiple sets of production batch data in bulk, check if retrieval results for different batches are isolated, and verify that the dataset binding logic is correct.
The values provided are common starting points and should be measured against the reader's own samples.