Data to be processed

Jump to navigation Jump to search
aiCAMstir logo
The aiCAMstir project investigates Artificial Intelligence based Computer Aided Manufacturing of Friction Stir Welds

Data to be processed in the aiCAMstir project are parameters, variables, boundary values and quality.

Key data to be processed

Three overlapping ellipses in three different colors show parameters, variables, boundary values and the overlap region represents the weld quality. Overlap region between variables and boundry values shows gap between the parts and tool wear (as shown in the table below). The overlap bewtween Parameters and Boundary values shows four entries, which are listed under boundary levels in the table. Three entries are shown in the overlap region between parameters and variables: Tool posiiton, tool forces and spindle torque, because it isn't clear which are the set and which are the resulting values
Data to be processed in the aiCAMstir project are parameters, variables, boundary values and quality.

According to digitalisation and data processing experts, the following categorisation of key data could be considered:

Parameters Variables Boundry Values Quality
Values that you set Values that result during welding Values that affect the results Quality
  • Rotation speed
  • Welding speed
  • Tool design
  • Tool temperature
  • Clamping forces
  • Workpiece materials
  • Workpiece cleanliness
  • Tool materials and coatings
  • Temperature of the workpieces
  • Temperature of the fixture
  • Geometry of the set-up
  • Spindle tilt angle
  • Tensile strength
  • Elongation
  • Visual appearance
  • Fatigue strength
  • Voids (pores)
  • Weld depth
  • Lack of penetration
  • Bend test results
  • Inclusions
  • Helium-tightness
  • Tool position
  • Tool forces
  • Spindle torque
  • Gap between the parts
  • Tool wear

The categorisation depends on the concept and complexity of the machine. Modern FSW machines provide close-loop force and position control algorithms, which complicate the distinction between parameters and variables. Some machines use the temperature as a parameter that can be set, e.g. to change the parameters if the tool gets too hot.

Please feel free to improve the classification above or to propose an alternative classification. Or use the discussion page for discussing changes.

Most important values for CNC programming and CFD modelling

A text box, a table and a chart showing by a formula, by numeric values and in the charts that the welding speed depends on worpiece material, tool design and workpiece thickness
Welding speed of FSW depending on material factor φFSW, tool factor ψFSW and thickness as originally presented by Stephan Kallee et al at INALCO '98[1]

Above mentioned values, which are often used in Welding Procedure Specifications (WPSs), are based on the parameter settings of conventional milling machines. Modern CNC machines, are controlled quite differently: Here the position in x-y-z-direction, the force in x-y-z-direction and the the spindle torque are the important values, although they cannot be measured directly, e.g. in articlulated arm robots and parallel kinematics robots.

On a conventional milling machine the the rotation speed and welding speed will be set, mainly depending on the pin length, the pin diameter and the material to be welded. If you plot the weld strength depeding on rotation speed and welding speed in an x-y-diagram to determine the parmeter envelope, you will notice that you get parallelograms instead of rectangles to describe the most useful parameter settings. Therefore, the term "tool forward movement per revolution" was introduced, which however still depends on the diameter of the pin chosen. In Japan the term "welding factor" is used particularly by those FSW users who use the term "FS double" instead of "FS double-U" for FSW. Experienced CFD modellers use a dimensionless factor, which is calculated from rotation speed, welding speed and pin diameter. Conical tool pins are therefore more complicated to be modelled.

The parameter setting could be simplified by using at least two dimension-less factors depending on the material (7000 series aluminium requires different settings to 6000 series aluminium) and tool design (a triflute tool welds faster than a cylindrical pin). The Greek characters φFSW and ψFSW have been introduced for the material factor and tool factor repectively, but not (yet) adopted by the FSW community. A third factor, depending on rotation speed, welding speed and pin diameter might be useful for comparing the results of welds made with different parameter seetings.


One technician controls the plunge depth on the FSW machine and three colleagues log the data
Friction stir welding data monitoring at NASA MSFC (Marshall Space Flight Center) in a video by NASA-MSFC, 2013
  • Computer Aided Manufacturing (CAM) includes data monitoring, data processing and data visualisation
  • Friction Stir Welding (FSW) is a special process, and its quality can only be determined statistically


  • Simplify the data exchange
  • Reduce the reaction time
  • Automate the documentation
  • Provide more attractive products and services[2]


  • CAM has no value by itself
  • Humans need to be involved
  • Raw data aren't providing information
  • Visual management is required[2]

The two approaches to collecting and processing of data

A photomontage of what a whole iceberg might look like. The upper part is a real image. The part below the waterline is another iceberg image upside down.
Most of the data will never be used

Bottom Up

  • Collect and store as much data as possible at all times
  • Use Data Mining to process the data according to specific needs
    • Advantages: Infrastructure will be installed. Data transfer is standardised
    • Disadvantages: There might be no short-term benefit. Creation and storage of data garbage(WP)[2]

Top Down

  • Specific objectives
  • Data collection is related to the needs
    • Advantage: Very good cost-benefit-ratio. Simple entry and fast progress
    • Disadvantages: Risk of insular solutions. Data might not be sufficient for the next project[2]

Data collection

Most FSW machines have a computerised data monitoring system for collecting data. The following data are most commonly collated:

  • x, y, z values of the CNC machine
  • Rotation speed
  • Welding speed, which is defined by the defined as the x, y, z value change with respect to time using the methods of analytical mechanics(WP)
  • etc

Data labeling

In most cases tabular data and images are being processed. A process has to be established to collate and process the data. This could be a skript or small project, which has to be created according to the specific needs of each application, i.e. to accomodate the format, in which a commercially available FSW machine stores the data. The aiCAMstir project participants can provide you with more information on sharing data in the most convenient format.

60 % of the work on applying artificial intellligence is often related to preparing the data. Subsequent steps will then be selecting the model and training the model.

We want to generate a standardised table, which will be used by the aiCAMstir users, to describe the parameters used during friction stir welding. This table weil help to address the following topics:

  • Cleaning
  • Visualisation
    • Cluster
    • Navigate
    • Purify[3]
  • Scaling
  • Formating
  • Verification
  • Correction
  • Selection
  • Balancing[4]

A suitable data input mask(WP) could be Microsoft Excel or Microsoft Access suitable, e.g. to check already in Excel sheet, whether the data entered are in a sensible range. These files are compatible with Microsoft Azure(WP) Machine Learning Studio .

Methods from concept to model

Academic approach

If you want to create a model of a friction stir welding you can progress academically. This is for instance useful if you are not sure about the types of data to be collected and their availability and quality.

  • Advantages: neutral, analytic and powerful
  • Disadvantages: large effort ist required

An ten-step example is as follows:

  1. Define the process and the objective of the model
  2. Select the data to be collected
  3. Determine the data field
  4. Find the data source
  5. Create an experimental plan (e.b. by design of experiments)
  6. Generate data
  7. Select a suitable model and build it
  8. Train the model
  9. Evaluate the model
  10. Assess the model: Is it okay?[2]

Pragmatic approach

Three technicians look at the control screens of an FSW machine while one of their colleagues controls the plunge depth with a portable human-machine-interface
Friction stir welding data monitoring at NASA MSFC (Marshall Space Flight Center) in a video by NASA-MSFC, 2013

The pragmatic approach follows the conventional development procedures. It makes use of expicit or intuitive expert knowledge. If large numbers of experts are involved this is occasionally called swarm intelligence. Their know-how needs to be transferred into the data processing system.

  • Advantages: Useful concepts can be developed in a short time frame. Already known correlations can be used. Quick validation.
  • Disadvanatages: not fully neutral. Depends on individual expertise. Risk of tunnel vision.[2]


  • Spider diagrams are easy to understand
  • KPIs (key performance indicators) should be used
    • Present operating and idle time, brutto capacity, number of rejected parts and overall efficiency etc.
  • Data monitoring should be applied
    • Create chart of the production rate, freak values, number of rejected parts, reasons for rejects including metallographic samples etc
  • Report
    • Trend of the reject rate
    • Monthly, quarterly or yearly production
    • Performance development[2]


  1. Stephan Kallee, Dave Nicholas, Haydn Powell and John Lawrence: Knowledge-base software package for friction stir welding. In: The proceedings of the 7th INALCO conference which was held at TWI, Cambridge in April 1998 - Joints in Aluminium - INALCO '98: Seventh International Conference, Woodhead Publishing, 14 October 1999.
  2. 2.0 2.1 2.2 2.3 2.4 2.5 2.6 Marion Purrio and Guido Buchholz (FEF Forschungs- und Entwicklungsgesellschaft Fügetechnik GmbH): Webinar: Vorgehensweise bei Digitalisierungsprojekten in der Schweißtechnik. 29 April 2021.
  3. cadGraph - Visualize your 3D CAD data, like never done before...
  4. Tobias Schulze and Konrad Riedel (MDesign): KI-Anwendung im MaschinenbauAnwendungsmöglichkeiten von Künstlicher Intelligenz und Maschinellem Lernen. Webinar on 9 September 2021.