A Brief Look at Methods & Technology

"We Don't Need the Ability to Search Something...,

We Need the Ability to Find Something!"

Dana Nickerson, Director of Global Process Management, Whirlpool

Accelerating Innovation with New Data Strategies, TDWI

That's a pretty clever way to speak about the poor accuracy of CAD database search tools, the majority of which are features of PLM systems.   

Attribute systems are based on tagging each model with the right keywords..., and are foiled by:

  • Reliance on humans to apply the correct attributes,

  • Synonyms,

  • Complex and easily forgotten naming rules, and

  • Global languages

Parametric systems were supposed to solve the weaknesses of attribute methods by using parameters such as surface area, volume and so on, as the basis for matching parts..., but are foiled by:

  • Different types of parts coincidentally having similar parametric data,

  • Similar parts having very different parametric data, and

  • The unavailability of data for many CAD formats including neutral formats and assemblies

In most cases, the result is typically either empty searches, or searches that overwhelm the user with clutter, which lead to abandoned searches and a reinforcement of a start-from-scratch culture that undermines part reuse or other use cases that search was supposed to be a solution for. 

Yet most engineering departments are still limited to out-of-date parametric search tools that produce either thousands of hits or no hits at all. 

Accelerating Innovation with New Data Strategies, TDWI


A New Approach

CADSEEK avoids the trade-off between omission and clutter errors with a technology that is capable of capturing the model's actual geometry (shape).  Instead of trying to simplify shape with a keyword or parametric profile CADSEEK precisely captures the full, rich geometry of the model at the high resolution of CAD, providing unprecedented accuracy and clutter-free results. 

Geometric Fingerprint

54714669852341297555487 ...


In essence, CADSEEK fingerprints each model so that the full geometry of each model is transformed into a very long string of numbers, so that the fingerprint of any model is instantly comparable for similarity to all others. 

Accuracy is undoubtedly the primary criteria used to judge a shape-search application, but it's really important to understand that there is a tremendous difference between identifying identical models and those that are similar.

Similarity is much, much harder to accurately detect, and the ability to not only match similarity but to also accurately determine the degree of similarity is the key difference between CADSEEK and all other search application. 

After all, if two models have identical surface area, there is a pretty good chance they are identical.  But similar models can be different in an endless number of ways which no number of parameters can account for.  And even if they could, not all differences affect parameters. 

Example #1

A prospective customer copied a CAD model of an ejection pin, and offset the head from the axis by a small fraction of a millimeter, leaving surface area and volume parameters of the two models identical.  CADSEEK was able to see the difference and scored the models as 99.97% similar. 

Example #2

The same customer believed two pins to be identical when CADSEEK reported a slight degree of difference.  On inspection it was true that the parametric values of the pins were identical, but examination of the models revealed different character inscriptions modeled into the underside of the head of each pin.  

Example #3

The models at right are obviously similar, but with significantly different parametric values.   No practical setting for tolerance in parametric values can match these types of similar models without an overwhelming number of coincidental clutter.

Example #4

CADSEEK's ability to be comprehensive while still controlling clutter is shown in the example below, where, in a real-world database of more than 20,000 models, a search was expanded, and then expanded again, becoming more comprehensive while still controlling clutter.    

Standard Search

Expanded Search

Exploratory 'Rough Sketch' Search

Example #5

Sample Analytics Report

Please take a look at a sample Analytics Report which shows CADSEEK's ability to determine similarity across an entire dataset with tremendous accuracy. 



Beyond accuracy..., there are quite a few other performance features and advantages of the CADSEEK technology:


With it's pattern recognition approach, CADSEEK has the intelligence to see the similarity of models with different scale, different feature location, similar but not identical aspect ratios, different internal geometry, left and right hand versions of parts and even parts with inverse shapes.  This intelligence allows CADSEEK to not just identify identical models but also provide meaningful similarity scores.


Obviously a search application can't be comprehensive if it can't search all relevant CAD formats.  CADSEEK can encode all types of proprietary CAD models, CAD assemblies, neutral CAD formats (STEP, STL or IGES), 3D laser scans, and even 2D images.

In addition, CADSEEK shape codes can also be embedded in QR Codes (quick read) and RFID tags for time-critical supply situations where the ability to find highly similar parts for potential substitution is needed (click for info on ShapeID).      


Many shape-search applications require a pair-wise comparison of each pair of models before a search using a new model as the search target can be executed.  The lengthy pair-wise comparison process means that there will be a very long delay before search results are available.  CADSEEK can encode a new model on the fly and retrieve search results in most cases in under two seconds.  More information can be found here on slides 14 - 17.  


CADSEEK has been successfully deployed in million-plus model datasets with more than 2,500 users, multiple PLM servers and proprietary data structures. 


CADSEEK is routinely integrated into CAD packages and PDM/PLM systems for efficient workflow.  Links to the PLM object pages are provided in CADSEEK search results allowing convenient and appropriate checkout of any model found in a search.  


CADSEEK can consider different facets of shape independently, allowing a search to be expanded in meaningful ways without overwhelming the user with irrelevant search results. 


CADSEEK technology is embedded in a integrated suite of tools spanning the enterprise from engineering design to field services with solutions for shape-based:

  • Search 

  • Identification and Substitution

  • Classification

  • Analytics


CADSEEK searches retrieve search results in less than two seconds even in large datasets. 


CADSEEK can be provided in a Command Line version allowing search to be embedded in proprietary engineering portals and tool suites.  


After the initial encoding process is complete CADSEEK not longer requires access to CAD models.  While a CADSEEK shape index will normally be kept behind an enterprise firewall, it has been proven that a CADSEEK fingerprint cannot be reverted to any kind of meaningful data, even by iSEEK.  This allows the potential to put search capability outside the firewall for customer access if desired.


CADSEEK products use a client/server architecture.  The process of encoding files averages less than 1 second per model, and for extremely large datasets the encoding process can be performed in parallel. 

A Historical Look at Shape-Search

Prior to the development of CADSEEK, shape search tools didn't actually capture shape directly from CAD models.  Geometric shape was simply too complex.  So instead, shape-search tools just simplified the problem by utilizing a surrogate, such as surface area or shadows cast from the model to represent the geometry.  There is some reasonable logic, after all, that if two models have the same surface area they might indeed be identical.   

Figure 1

Figure 2

Figure 3

But a much more difficult challenge than identifying identical parts is matching parts for similarity, and also scoring the degree of similarity.  This capability enables many use cases and is the key advantage of CADSEEK technology.

A Complex Challenge

Capturing and comparing shape is complex because geometry is composed of numerous facets, and also because of the complicating factors created by CAD software, including orientation and translation.


  • Topology (adjacency of elements)

  • Dimensions (scale)

  • Constraints (type of adjacency)

  • Features and Feature Location

  • Internal Geometry

All combined = Geometry


  • Orientation of the CAD model

  • Translation of the CAD model

  • Proprietary format of CAD models

  • Models vs. Assemblies

Translation = location of model in CAD Volume

Surrogate Methods

Surrogate methods were initially intended to spot potential duplicate models, at a time when the ability to do so was a significant breakthrough.  The concepts behind the surrogate methods are pretty easy to understand, but the related pros and cons are not so obvious. These include challenges with accuracy (omission and clutter errors), inability to encode assemblies, being limited to a single CAD format, and the inability to immediately perform a search with a new model. 


Four surrogate methods are profiled below.  CADSEEK is the only shape-search application in the market that is not based on one of these methods. 

Parametric Substitution


Graph Matching


Parametric substitution compares one or more parametric values, such as surface area, with the hope that coincidences  indicate potential identical models. 

Projections (shadows) of each model are made from multiple perspectives (to account for orientation differences).  Algorithms then try to align the shadows to find matches.  

Model elements such as edges and loops are graphed, and the graphs are compared.  A key challenge is that even a small amount of complexity leads to an NP complete situation which cannot be solved. 

The CAD volume is partitioned into smaller and smaller cubes and a parametric value such as surface area is 'mapped' for each cube.  The maps are then compared to find potential matches. 

Simplification = Trade Offs

Surrogate methods are based on simplification, and simplification isn't all that bad when limited to searching for identical models, but there are a couple of critical weaknesses when it comes to searching for similarity. 

There is no rational logic to support the a belief that a 93% similarity in surface area should correlate to 93% similarity in shape, or any similarity at all for that matter.

Moreover..., two parts can have a 40% or even 400% variance in surface area and still be highly similar, but if such a large tolerance were added to the search the coincidental matches would be overwhelming.   

This trade-off of clutter vs. comprehensiveness is a dilemma for each of the surrogate methods. 

In theory, parametric techniques can gain accuracy if multiple parametric measures are used in concert, such as combining surface area, volume, moment of inertia and file size, but there are several practical limitations. 


First, there are significant challenges in creating an reliable method that works across all types of parts from bolts to manifolds to landing gear assemblies.  The idea commonly used is that, if surface area changed, but two parts still have highly similar moment of inertia parameters, then the two models might be similar.  The reality is that most differences between similar parts, such as a change in a hole location and size, will materially alter each and every parametric value.  In addition, two parts can differ in scale by 40% or 400% and still be very similar, but those major differences in parameters will undoubtedly eliminate those different sized yet similar models from the search results.  If somehow they are not eliminated , they they will will typically be hidden in an overwhelming amount of clutter due to the loose search tolerances. 

Second, the parametric data has to be available, and that includes across all CAD formats, neutral formats like STL, and especially in assemblies. 

Yet most engineering departments are still limited to out-of-date parametric search tools that produce either thousands of hits or no hits at all. 

Accelerating Innovation with New Data Strategies, TDWI

A Better Recipie

So why is CADSEEK any better?  CADSEEK is able to be both accurate and comprehensive because it is the only shape-search technology that considers all facets of shape using surface-based methods like modern CAD software, allowing CADSEEK to operate at the same resolution as the CAD model. 

This also gives CADSEEK the unique ability to selectively ignore one or more facets of shape when desired.  When a method only considers a single facet of shape, it can't afford to ignore any data.  But CADSEEK can, for example, put more emphasis on the general shape, and less emphasis on features and scale.  The ability to change the emphasis of a search allows CADSEEK to provide meaningful search tiers that increase comprehensiveness, but still control clutter. 

We would appreciate the opportunity to demonstrate CADSEEK's ability during a live demonstration session.  Please click here to request a demonstration. 


Documents Library

PowerPoints (PPTs) Work best with Edge or Chrome browsers  

iSEEK Brochure

Insights for Evaluating a Shape-Search Application

Sample Analytics Report

White Paper - Analytics for Engineering Data

White Paper - Economics and Benefits of Shape-Search

Blog Post - Encouragement is Not Enough to Drive Part Reuse Initiatives

Blog Post - Can a Live-Search Feature in CAD Help Drive Part Reuse