What distribution should I expect a tensile strength test to fit?

M

Matthew_Hopkins

#1
Dear friends,
Please advise a pondering quality engineer with this perhaps simple question.

We are doing tensile strength tests of soldered joints on a steel wire.
When I perform different tests of normality, the results are indicating normality but not with a certainty I'm satisfied with. Since the tests are destructive and expensive it's hard to achieve extensive amount of data to base conclusions on.

May I expect the results from a tensile test to be normally distributed or something else, like Weibull? Logically there is an absolute possible Min. (e.g. zero) and Max. value for a strength distribution. Does a normal distrubition requires a probability >0 for any value or how do I handle a discontinued distrubition? Is it even possible to determine an "absolute Min."?

Thanks for any input in advance,
Matthew
 
Elsmar Forum Sponsor
#2
Matthew_Hopkins said:
May I expect the results from a tensile test to be normally distributed or something else, like Weibull?
We perform tensile testing on every single product (Stainless steel coils) which means that we have an enormous amount of data to choose from, and our results are definitely normally distributed. Obviously, the lower end of our distribution curves are nowhere near 0.

Can you provide us with more info about why you suspect that your products tensile strength may not be normally didtributed? Perhaps I'm missing a beat here, but offhand I have a hard time imagining why it would not be?

Btw: Based in Sweden, are you? :bigwave: Whereabouts?

/Claes
 
M

Matthew_Hopkins

#3
Claes Gefvenberg said:
Can you provide us with more info about why you suspect that your products tensile strength may not be normally didtributed? Perhaps I'm missing a beat here, but offhand I have a hard time imagining why it would not be?
My main reasons to be suspicious are the two peaks in the histogram, and S-shaped form in the normal distribution plot. Secondly, it fails other normality test (e.g. Shapiro-Wilks) on a certain confidence level. Third, the fracture location may vary between the actual joint and the wire material (however not corresponding to the two histogram peaks). Fourth, different operators/machines are performing the joint operation. Categorizing based on operator gives better normality fit but reduces the available sample size and confidence even further (i.e. from a total N of 50 to 10 per operator). As well I'm starting to suspect some flaw in the test method, like an unstable fixture affecting the results. The test in question is trickier than it may sound. Obviously, we have to make sure that the tensile strength does not fall below a certain level. Theoretically, a normal distribution gives a certain probability for a random unit to have the value zero, or is this just a major logical error in thinking?

Claes Gefvenberg said:
Btw: Based in Sweden, are you? :bigwave: Whereabouts?
That's true, Uppsala, Sweden, nice to see other representatives around :agree1:

//Matthew
 
#4
Matthew_Hopkins said:
Third, the fracture location may vary between the actual joint and the wire material (however not corresponding to the two histogram peaks).
That complicates matters. That should produce two normal distributions merged into one that is not. I cannot explain, however, why they do not correspond to the two peaks :confused: .

Matthew_Hopkins said:
Fourth, different operators/machines are performing the joint operation.
Which means several parallel processes? If so, I think you will have to evaluate the different processes separately, or you will again see several normal distributions merged into one.

Matthew_Hopkins said:
That's true, Uppsala, Sweden, nice to see other representatives around
Likewise :agree: .

/Claes
 
M

Michael Walmsley

#5
Send me the data. I'll run it through several analytical packages I have.

Mail addie removed in order to prevent bots from harvesting it and sendin spam your way. People can still contact you by clicking you name in the top left corner of the post and selecting mail or PM from the menu found there.

/Claes


Generally, data distributions that fit the tensile arena fall into the

Extreme Value category.

Mike
 
Last edited by a moderator:
M

Matthew_Hopkins

#6
Claes, you're probably right. Unfortunately the operator variation may be significant and the amount of data is most likely insufficient to determine if there actually are two distributions due to different fracture locations. Looks like I will end up with further destructive testing in this matter.

Michael, I appreciate your offer very much and will of course send you the data. As it is, the result tells me nothing, a second opinion would be great.

//Matthew
 
D

Dave Dunn

#7
Matthew_Hopkins said:
Claes, you're probably right. Unfortunately the operator variation may be significant and the amount of data is most likely insufficient to determine if there actually are two distributions due to different fracture locations. Looks like I will end up with further destructive testing in this matter.

Michael, I appreciate your offer very much and will of course send you the data. As it is, the result tells me nothing, a second opinion would be great.

//Matthew
I have to agree with Claes; combining data from two different processes, even if they're designed to be the same, will frequently give you a bi-modal distribution due to variance between the processes. It's the same as measuring a dimension on multiple cavities from the same molding tool in my work and expecting the distributions to be the same. The tooling is designed to be the same, or at least similar, but variances will show themselves.

This is not to say that you can't combine such data, but it will make it more difficult to determine what results each of the processes is actually producing.
 
M

Michael Walmsley

#8
Attached are the analysis results.

(1) I used Weibull Analysis , specifically in the area of assessing competing modes (risks) and their potential impact on your results.

I looked at it in terms of assessing results for :

Wire Type (X Y)
and later Operator impact (A B C D E).

Both affect your results when looking at the multiple plots .

(2) A principle components analysis (NIPALS) and General Regression Modeling analysis were then undertaken.

In both cases , Operator , then Wire type , in that order were listed as having an impact on your results. The model
looks to be non-linear in nature (eg the correlation coefficient is low indicating a non-linear model).

Though they showed up as borderline insignificant at alpha = .05 level , it is still close enough to generate a
suggestion:

Conduct an MSA to assess true nature of operator and wire impacts on your measurements. From here you can
at least resolve some of the variability issues in your measurement system.

Hope this helps

Mike
 

Attachments

Hershal

Metrologist-Auditor
Staff member
Super Moderator
#10
Similar tensile tests in the U.S. market, under ASTM standards for metal and rubber coupons (as they are known) are typically considered normal distribution.

Test labs here that are running those tests use the numbers in their uncertainty calculations.

Hope that helps a little.

Hershal
 
Thread starter Similar threads Forum Replies Date
M Should I use binomial or Poisson distribution formula? Statistical Analysis Tools, Techniques and SPC 6
A Can anyone share a Distribution Agreement template under MDR 2017/745? EU Medical Device Regulations 0
P Sample Size for Distribution Simulation Testing Inspection, Prints (Drawings), Testing, Sampling and Related Topics 11
D Private Label Distribution - US Other US Medical Device Regulations 1
D Question regarding "storage and distribution" ISO 13485:2016 - Medical Device Quality Management Systems 1
M Proof of distribution - FedEx/UPS tracking numbers Manufacturing and Related Processes 2
O Are power distribution units Multiple Socket Outlets (MSOs)? IEC 60601 - Medical Electrical Equipment Safety Standards Series 1
JoCam Medical Device Distribution in the UK EU Medical Device Regulations 6
M Prescription medical device distribution licenses in the U.S US Food and Drug Administration (FDA) 3
I Control of Documentation Distribution - Document Control ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 2
M US distribution Medical Device Medical Device and FDA Regulations and Standards News 8
0 Interesting Discussion Analysis of half normal distribution in minitab Using Minitab Software 11
Marco Bernardi Finding a flat or not modeling distribution: how to manage it? Capability, Accuracy and Stability - Processes, Machines, etc. 24
T Was Just told we could ONLY buy material for AS9100 customers if the DISTRIBUTION house is AS9100 / AS9120 AS9100, IAQG, NADCAP and Aerospace related Standards and Requirements 3
P Medical Device not on commercial distribution but installed in the establishment. 510k? Other US Medical Device Regulations 10
M How do you Audit an Distribution Facility according to MAQMSR IATF 16949 - Automotive Quality Systems Standard 3
K Inspection Certificate EN 10204 2.1 - Distribution company Other ISO and International Standards and European Regulations 4
S Medical Device Software Distribution for a CE-marked medical device via a USB memory stick EU Medical Device Regulations 8
L How to evaluate the process capability of a data set that is non-normal (cannot be transformed and does not fit any known distribution)? Capability, Accuracy and Stability - Processes, Machines, etc. 12
K Distribution medical devices to US Embassies 21 CFR Part 820 - US FDA Quality System Regulations (QSR) 2
P AS9120B Control of External Providers for Franchised Distribution AS9100, IAQG, NADCAP and Aerospace related Standards and Requirements 0
C IATF 16949 - RSL - Distribution Center Clasification IATF 16949 - Automotive Quality Systems Standard 3
S Ways to attract clients - Flyer distribution in Toronto, Canada Service Industry Specific Topics 1
N Apply control limits to a non-normal distribution Statistical Analysis Tools, Techniques and SPC 13
G Plotting capability study data - I have a bimodular data distribution Capability, Accuracy and Stability - Processes, Machines, etc. 4
Sidney Vianna ISO 13485 Certificates Distribution around the World - 2018 ISO 13485:2016 - Medical Device Quality Management Systems 4
F Definition Commercial Distribution - Definition - Scope of Commercial Distribution Definitions, Acronyms, Abbreviations and Interpretations Listed Alphabetically 2
R Non-normal Distribution Selection where the system is constantly being corrected Capability, Accuracy and Stability - Processes, Machines, etc. 11
J Distribution and PPAPs for products they do not manufacture APQP and PPAP 3
G Uncertainty Experts - ISO 17025 - If the Distribution is Rectangular Measurement Uncertainty (MU) 2
D Distribution of a Class 3 Medical Device EU Medical Device Regulations 4
K Distribution Centers - NSF Audit Qualification and Validation (including 21 CFR Part 11) 1
A Not all characteristics follow a Normal Distribution - How do you do SPC Chart Capability, Accuracy and Stability - Processes, Machines, etc. 5
F Distribution Records required by FDA - 21CFR820.160 21 CFR Part 820 - US FDA Quality System Regulations (QSR) 2
V Process Capability for parameters with non-normal distribution. Capability, Accuracy and Stability - Processes, Machines, etc. 16
P IVD device distribution, what to do? Japan Medical Device Regulations 1
M Medical Device Distribution in Canada - Do we need both MDEL and MDL? Canada Medical Device Regulations 8
E Medical Device Distribution Quality Management in South Korea Other Medical Device Regulations World-Wide 0
D Calculating Cpk on Non-Normal Data Distribution Capability, Accuracy and Stability - Processes, Machines, etc. 10
L Choosing the correct Distribution for Acceptance Sampling Inspection, Prints (Drawings), Testing, Sampling and Related Topics 19
A Medical Device Distribution in the EU to Patients (End-Users) EU Medical Device Regulations 3
L Intellectual Property Protection - Storage, Access, Distribution, etc ISO 9000, ISO 9001, and ISO 9004 Quality Management Systems Standards 6
J Non-Normal Distribution Data - Tolerance Intervals and Minitab Using Minitab Software 7
J Lean Appropriate KPIs for a Distribution Centre Lean in Manufacturing and Service Industries 1
C Measurement System Artificially Creates a Skewed Distribution Inspection, Prints (Drawings), Testing, Sampling and Related Topics 16
J Lean Transformation Project - Distribution centre in Europe Lean in Manufacturing and Service Industries 1
L To share or not to share? Work distribution when someone goes on vacation Quality Manager and Management Related Issues 19
J Capability Analysis - Unusual Statistical Distribution of my Proccess Capability, Accuracy and Stability - Processes, Machines, etc. 5
S Distribution of IVD products for RUO and IUO-FDA Final Guidance 21 CFR Part 820 - US FDA Quality System Regulations (QSR) 0
0 Medical Device Distribution Centre and Implementing ISO13485 ISO 13485:2016 - Medical Device Quality Management Systems 1

Similar threads

Top Bottom