Frequently Asked Questions
Operational and Architecture FAQs
Q: Does CADSEEK modify my models?
A: No. CADSEEK accesses each model to create a shape signature, which typically takes a few seconds per
model. After the encoding process is complete CADSEEK no longer requires access the original model unless it is modified.
Q: What happens if a model is modified, or a new file is created after the original encoding process?
A: Periodically (nightly or weekly) CADSEEK queries the file repository (PLM, PDM, folders, etc.) to identify files
that have been newly added or modified. Any modified or newly added files are encoded and their signatures are added to the search index. The process is completely automated and is typically run each evening, but the frequency can vary depending on the rate at which new files are added to the CAD database.
Q: How much load does CADSEEK place on the PLM or other file servers?
A: During the initial encoding process CADSEEK request models in a serial manner so that it does not
overwhelm the server and lock other users out. Once files are encoded, CADSEEK no longer requires access to the original CAD models. Searches occur on the CADSEEK server and therefore CADSEEK puts NO load on the PLM or other CAD server. No shape search product performs searches based on actual CAD models in real-time so CADSEEK is not unique in this manner.
Q: How long does the original encoding process take to create the search index?
A: This depends mostly on the number of CAD models and the encoding hardware. Typically a dataset of
100,000 models can be easily encoded within a few days using a single machine with a standard quad-core processor. Server class hardware will significantly shorten the process. For extremely large datasets the encoding process can be run in parallel on multiple machines.
Q: Does CADSEEK protect my intellectual property?
A: Yes. Our customers have confirmed that it is mathematically impossible to invert a CADSEEK shape
signature into any kind of meaningful data, even by an iSEEK employee.
Q: What are the CADSEEK hardware requirements for operations?
A: CADSEEK is configured in a client/server Windows architecture, and can be run in a virtual environment.
The server requirements are far below that of a PLM. RAM requirements vary from 8 to 32 GB. The number of computing cores is highly dependent on the number of new models being encoded, i.e., if real-time quotes are being developed by matching files uploaded by customers, then a greater than typical amount of encoding will be occurring, which increases the number of required cores.
One or more CADSEEK servers can be run in parallel, but one server is the norm. John Deere, for example, has a shape index of over one million parts and provides service to more than 2,500 users across 26 global locations all from a single CADSEEK server. Client machines are generally described as mid-range including laptops. Dedicated graphics cards are recommended. Any computer that can run CAD software can easily run CADSEEK.
Q: What file types and infrastructure can CADSEEK be deployed in?
A: CADSEEK can encode any type of popular CAD formats for both models and assemblies; neutral files such as STEP,
STL and IGES; and other digital 3D formats produced by laser scanning or point-cloud devices. Integration is available for popular CAD packages such as Creo or Solidworks; PDM or PLM systems such as Windchill or Teamcenter; enterprise software such as SAP; and also proprietary databases.
Q: How long is the typical deployment timeline?
A: In general, the timeline typically ranges from a few days to a few weeks. Multiple PLM server
configurations and proprietary databases will require additional time. The first step is to create integration with the CAD file repository. Automated processes encode models to create a CADSEEK shape signature for each model and assembly. Some files will not encode due to either missing dependency files, or due to incomplete geometry. CADSEEK creates a missing dependencies report to aid the process. Once the encoding process is complete an automated classification process groups models into families. Meta-data from the PLM, spreadsheets or other sources is then imported into CADSEEK for text based search and filtering.
Performance Related FAQs
Q: Why Shouldn't we just focus on improving attributes?
A: Attributes strategies require planning for naming rules, and the designer of the model, or a buyer of a
part, becomes responsible for applying attributes so that they and others can find the model in the future. Many companies put significant time and expense into developing attribute naming schema. The limitation with text-based searching, however, is that a keyword like 'bracket' won't be comprehensive because some designers will instead apply a synonym such as strut, brace or support; use a different language like soporte, 托架, klammer, or ブラケット; or forget to apply an attribute tag altogether.
Additionally, an attribute like valve can apply to models with such little similarity that the keyword produces hundreds of meaningless results..., but adding additional descriptors becomes too complicated to ensure universal usage.
But every model has shape, automatically. Shape is not subject to synonyms, different languages or the error of not bothering to apply an attribute. This makes shape automatically comprehensive, universal and impossible to omit from the PLM record.
Attributes are simply unreliable, and globalization is exacerbating the issue.
Q: What parametric data is needed for CADSEEK to function?
A: CADSEEK requires NO parametric or attribute data to function.
Our CADSEEK technology is instead able to capture the full geometry of the model in high resolution. However, iSEEK does normally include parametric data and any other attributes a customer desires in the CADSEEK search index. This allows a search to be started with text if desired, or allows text or numerical values to be applied as search filters.
Q: What is the practical value of the Navigator module in CADSEEK Polaris?
A: To start a search based on shape, you first have to find a target - essentially a search is telling CADSEEK
to find more like this. The Navigator makes finding the first model efficient and convenient by arranging the entire CAD database into shape-based families (a classification by shape). The Navigator displays each family as a tile in the 3-dimensional space of the Navigator. The families are identified by a thumbnail icon using the most 'average' part in the family on the face of the tile, allowing visual browsing of the entire CAD database. Families are arranged in the Navigator module by aspect ratio, so a part that is a perfect sphere or perfect square would appear in the bottom right of the Navigator window, and a long skinny part would appear in the bottom left corner. Once a user spends a little time using the Navigator they begin to intuitively be able to locate parts based on their general shape. This makes the Navigator a highly efficient way to locate the initial search target and begin a shape search. But the Navigator is just one of ten search options in the CADSEEK suite of products.
Q: How is a family in the Navigator different than a search?
A: Families are meant for rapid visual browsing and the quick selection of the initial 'target', from which to
begin a more focused search. The models that are automatically selected for a families is based primarily on general shape and less on features and scale. For example, a family may be composed of pumps of various sizes with both single and dual output ports. Selecting a family member with a dual port as the starting point for a standard search will likely create search results of only dual port models in a fairly tight size range. In many situations, browsing the family may immediately lead to finding the desired model, and in that scenario, selecting the family is essentially the same thing as a search.
Q: How do you characterize the accuracy of CADSEEK?
A: CADseek operates at the resolution of CAD. Even differences in saving a CAD model in another format, or
tesselating the model will typically be detected by CADseek. Two examples follow:
-
A maker of extraction pins used in the injection molding business challenged a CADSEEK Analytics report that said two models were 99.97% similar, when instead the customer believed the models were identical due to their identical parametric values. Upon inspection of the models it was discovered that the designer had modeled inscribed tiny characters on the underside of the pin heads and therefore the models were indeed different.
-
A maker of cylindrical parts with multiple diameters along the shank copied a model and intentionally offset one of the cylindrical segments by a fraction of a millimeter. CADSEEK was able to detect the models as highly similar, but not identical even though the parametric values were identical.
A great way to see CADSEEK's ability to be both comprehensive and yet avoid clutter (accuracy) is by browsing a sample Analytics report, which shows page-after-page of what are essentially search results.
Q: What does a measure like 97% similarity mean?
A: It's really not possible to provide a good answer that isn't the size of a thesis. What can be said is that the
similarity score is a function comprising all facets of shape. Elements such as wall thickness, internal geometry, curvature, feature location and the like are all included in the shape fingerprint. A score of 100% means identical. A score of 98% is going to take some careful looking to spot the differences. 70% would have reasonable similarity. 50% .
Q: How long does it take to run an Analytics report in a dataset of 20,000 models?
A: Such a report, on a workstation class PC would take less that twenty minutes. Maybe less than ten. In
doing so, CADSEEK would have made 399,980,000 model comparisons (20,000 * 19,999), and would report the results down to a user-defined limit, i.e., show me all models that have 87% or greater similarity. If each comparison could be performed in one minute by a person, the same analysis would require over 500 years with the disadvantage of lacking a consistent and rational score for similarity.
Evaluation and Sales FAQs
Q: How is CADSEEK sold?
A: CADSEEK can either be purchased directly from iSEEK or via your favorite reseller of engineering
applications. Our sales process is pretty nimble.
-
A typical first step is for iSEEK to provide a web demonstration using generic data.
-
The web demo is followed by an invitation to demonstrated CADSEEK using proprietary data.
-
We encourage a volume of 1,000 to 10,000 models to ensure a good test.
-
The dataset will ideally contain both models and assemblies, of different formats, with some models duplicated in neutral formats such as STEP or STL, a wide variety of model types, and some deep pockets of duplication and similarity. A key issue for evaluators is the ability to spot false negative results (similar models that should have been reported as matches but weren't). While false positive clutter is obvious, false negative omissions are invisible unless an evaluator knows what to expect in the search results (i.e., 'I wonder why part Z-72b wasn't including in the search results).
-
Q: What does CADSEEK cost to purchase and deploy?
A: There are quite a few variables involved in getting to a final price, such as integration requirements,
custom features and scope. Deployment can occur in a step-wise fashion for specific situations such as acquisition or data cleansing. Our Analytics reports showing duplication or similarity can also be provided as a service. Most importantly, several of our customers have formally studied the ROI CADSEEK is capable of generation including John Deere, MOEN and Embraer. All have determined that they recouped 100% of their investment in CADSEEK in well under twelve months. References can be provided.