Engineering Knowledge Management
SolidWorks PDM stores thousands of parts but engineers still can't find what they need. Here is exactly why text search fails, and what AI-powered geometric search changes in 2026.
·
⏱
7 min read
Michelle Ben-David
Mechanical Engineer · B.Sc. Technion
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
AI-powered fastener selection cross-checks geometry, material, thread engagement, torque requirements, and your approved vendor list in seconds — eliminating the back-and-forth that adds days to assembly design cycles.
An engineer needs a stainless flange nut with a specific head height. She checks SolidWorks PDM. Types "M8 stainless flange nut." Gets 847 results. Filters by project, by revision state, opens ten files. Nothing matches. Closes PDM, opens SolidWorks, and starts modeling the part from scratch.
The part she needed was result 23. The filename was "STD_HW_012_REV4." The description said "hardware item." PDM stored the part perfectly. It just could not surface it when it was needed.
This is not a rare edge case. In most engineering organizations running SolidWorks PDM, engineers fail to find existing parts several times per week. The workaround is always the same: design from scratch, add another duplicate to the vault, and repeat the cycle next quarter.
SolidWorks PDM was built on a reasonable assumption: engineers know what they are looking for well enough to describe it. The search function queries file names, custom properties, description fields, and revision states. It does not read geometry. It does not understand intent.
This works when every file in the vault is named consistently, every property card is filled out completely, and every engineer on every team follows the same conventions. In practice, in vaults that have been in use for more than three years, none of those conditions hold reliably.
File naming drifts as different engineers apply different conventions over different project phases. Custom property fields are filled out under deadline pressure with whatever is fastest. The description field contains whatever the engineer who created the file thought was most important at the moment of save.
The result is a search tool that is technically accurate but practically limited. It finds exactly what you searched for in exactly the field you searched. It cannot find the part that matches what you meant.
IN PRACTICE · HP ENGINEERING TEAM
"We had a senior engineer leave after 11 years. Within two weeks, the team was querying his documentation through Leo like he was still there. That's when we knew this was different."
— Senior Mechanical Engineering Manager, HP Inc.
The standard response to PDM findability problems is to add more structure: more required fields on the data card, more detailed naming conventions, more mandatory review steps before release. Engineering organizations have been trying this approach for twenty years.
It does not work because the problem is not a schema problem. It is a behavior problem.
Engineers are not going to spend three extra minutes completing a well-structured metadata form every time they save a part file. They are under schedule pressure. The design review is in two hours. The metadata gets the minimum necessary to pass validation and move on.
Mandatory fields make this worse in a specific way: engineers fill them out with placeholder values ("TBD," "standard," "N/A") to satisfy the form and keep moving. The field exists in the data card. The useful signal is not there.
Adding more required fields increases the noise. It does not make the vault more searchable. The only real fix is a search method that does not depend on metadata that was never reliably filled out.
There is a second layer to the problem. PDM stores files. It does not store the reasoning behind those files.
When a designer selects 316 stainless for a bracket instead of 304, that decision was made for a reason. A corrosion study from an earlier program. A supplier qualification change. A customer specification requiring marine-grade material. That reasoning exists somewhere: in a project report, a stress analysis document, an email thread, or in the memory of the engineer who made the call.
When that engineer leaves, the reasoning leaves with them. The file stays in PDM. A new engineer opens the same bracket, sees "316 stainless," and has three options: track down someone who might remember, run a new analysis to check whether the material choice matters for the new context, or leave it as-is and accept that they do not know why.
None of those are good engineering. All three are expensive.
This is the institutional knowledge problem layered on top of the search problem. The PDM vault contains the results of engineering decisions. It does not contain the decisions themselves.
An AI system that reads native CAD geometry changes what can be found and how it can be found.
Geometric similarity search reads the actual shape of parts from SLDPRT and SLDASM files and returns matches based on how geometrically similar they are to the search input. That M8 flange nut with the specific head height can be found by showing the system a matching geometry, regardless of what the file was named or what the property card contains. Results include dimensional comparison data so engineers can see exactly how close each match is before opening anything.
Leo AI indexes SolidWorks PDM vaults by reading the native files directly — SLDPRT, SLDASM, STEP, IGES, and others. When an engineer describes what they need in plain language ("M8 stainless flange nut, head height under 6mm, released state, marine-grade material"), Leo searches against what the parts actually are geometrically, not what they were called. The 847-result query that returned nothing useful becomes a four-result query with dimensional match data. The engineer finds the right part in under two minutes.
Beyond findability, Leo indexes the organization's full knowledge base: design reports, stress analyses, supplier qualifications, archived project documentation, meeting notes. When an engineer asks why a bracket is 316 stainless and not 304, Leo returns the original corrosion study, the material justification from the project report, and the design note from the revision that made the material change. The institutional knowledge does not disappear when engineers leave. It becomes retrievable.
Can AI fastener selection handle custom or non-standard fasteners?
What is SolidWorks PDM used for?
Why can't engineers find parts in SolidWorks PDM?
How do I search SolidWorks PDM more effectively?
What is geometric similarity search in CAD?
Can AI search a SolidWorks PDM vault?
How does Leo AI integrate with SolidWorks PDM?
Consider an engineering organization with a 40,000-part SolidWorks PDM vault accumulated over 12 years. Studies on vaults of this age consistently show 20 to 30 percent part duplication: parts that are geometrically nearly identical but exist as separate items because the engineers who created them could not find the existing version in time.
Each duplicate part carries a procurement record, a revision history, and a drawing set. Each duplicate part has to be managed through Engineering Change Orders when standards or materials change. Each duplicate is carrying overhead that compounds over time.
Reducing duplication starts with making existing parts findable before engineers create new ones. An AI that reads the vault geometry and returns similarity scores against existing inventory gives engineers a reliable way to check whether a suitable part already exists. When engineers can find what is there, they stop adding what is not needed.
Leo integrates with SolidWorks PDM without replacing it. The PDM system continues to manage check-in, check-out, revision control, and workflow approvals. Leo sits on top of PDM as an intelligent search and knowledge layer, reading what is already there and making it accessible in ways that text-based search cannot.
"The connection to our PDM and using that as a data source is legit the best thing ever. I found three viable bracket options fitting my exact envelope constraints — in minutes, not days."
— Eytan S., R&D Engineer
SolidWorks PDM stores engineering history. The problem is never storage. It is retrieval.
Engineers fail to find existing parts not because the vault is disorganized, but because text-based search was designed for metadata that was never reliably filled out. The institutional knowledge layered on top of the geometry — why decisions were made, what alternatives were considered, what the constraints actually were — lives outside PDM entirely.
AI that reads native CAD geometry changes both problems at once. Geometric similarity search makes parts findable regardless of naming. An indexed knowledge base makes engineering rationale retrievable without tracking down the person who made the decision. The result is less duplication, faster decisions, and engineering context that survives team turnover.
SolidWorks PDM (Product Data Management) — A file management and version control system integrated with SolidWorks CAD. It stores part files, assemblies, drawings, and revision history in a central vault accessible to the entire engineering team.
Vault — The central repository in a PDM system where all CAD files and documents are stored, checked in, and version-controlled. Access is managed through user permissions and workflow states such as Work in Progress, Under Review, and Released.
Custom Properties — Metadata fields attached to SolidWorks part and assembly files (such as material, part number, description, and revision). These fields are the primary data source that PDM search queries against.
Data Card — The metadata form displayed in SolidWorks PDM when a file is viewed or saved. Engineers fill this in when creating or checking in parts. Incomplete or inconsistent data cards are the primary cause of poor search results in large vaults.
Geometric Similarity Search — A search method that compares the actual 3D geometry of CAD parts and returns results based on shape and dimensional similarity rather than text metadata. Enables finding existing parts that match design intent even when file names and property fields are incomplete.
PDM Vault Duplication — The accumulation of near-identical parts as separate vault items because engineers could not find the existing version at the time of creation. Increases procurement cost, BOM complexity, and revision management overhead.
Tribal Knowledge — Engineering expertise, design rationale, and decision history that exists in people's heads or informal documents rather than in formal systems. Lost when experienced engineers leave the organization.
STOP LOSING ENGINEERING KNOWLEDGE
Every resignation email is a knowledge transfer you are not ready for.
Leo AI makes the institutional knowledge already in your CAD vault searchable by every engineer on your team - before the next senior engineer walks out the door.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by HP, Scania, GE and 50+ teams
