EP0359378A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Artificial intelligence for adaptive machining control of surface finish
Assignee: Ford-Werke GmbH
•
Filing: 1989-07-26
•
Granted: —
•
Expires: 2009-07-26
Abstract
Abstract A method of using a mathematical model to adaptively control surface roughness when machining a series of workpieces or segments by: (a) linearizing a geometrical surface roughness model; (b) initializing said model essentially as a function of feed; and (c) subjecting the initialized model to computerized estimation based on roughness and feed values taken from the last machined workpiece, thereby to determine the largest allowable feed for attaining a desired surface roughness in subsequently machined workpieces or segments of the series. The mathematical model is an algorithm of the form R = [1262.79]f²/r. The model is linearized and initialized to give the form R = β₁s R f+β₂R max where R is the actual roughness and R max is desired roughness, f is actual feed, β₁ and β₂ are coefficients to be updated by estimation, and s R is a scale factor chosen to make the first term have the same order of magnitude as the second term. Computerized estimation is carried out by converting the above initialized linear model to vector matrix notation with provision for the estimated coefficients in the form R n = ϑ T x n , where R n is measured roughness, x n is a computed vector taken from measured feed, and ϑ is a vector to be estimated with "T" denoting the transpose of the vector. After inserting roughness and feed values into such vector model, taken from the last machined workpiece, the coefficients are recursively estimated by sequential regression analysis.
Inventors
- 👤 Charles Laisau Wu
- 👤 Roger Kent Haboush
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-10-31 07:42
CPC Classifications
G; G05; G05B; G05B19/00; G05B19/02; G05B19/18; G05B19/416; G05B19/4163; G05B2219/00; G05B2219/30; G05B2219/33; G05B2219/33002; G05B2219/37; G05B2219/37402; Y; Y10; Y10S; Y10S706/00; Y10S706/902; Y10S706/903; Y10S706/904
Claims Analysis
Total Claims: 25
Claim 1
Independent Claim
91 words
A method of using a mathematical model to obtain adaptive feed control of a cutting tool for improving the surface finish of a series of machined workpieces or segments, comprising: (a) linearizing a geometrical model of surface roughness essentially as a function of feed; (b) initializing said model; and (c) subjecting the initialized model to computerized estimation based on roughness and feed values taken from the last machined workpiece, thereby to determine the largest allowable feed for attaining a desired surface roughness in subsequently machined workpieces or segments of the series.
Claim 2
Independent Claim
91 words
A method of using a mathematical model to obtain adaptive feed control of a cutting tool for improving the surface finish of a series of machined workpieces or segments, comprising: (a) linearizing a geometrical model of surface roughness essentially as a function of feed; (b) initializing said model; and (c) subjecting the initialized model to computerized estimation based on roughness and feed values taken from the last machined workpiece, thereby to determine the largest allowable feed for attaining a desired surface roughness in subsequently machined workpieces or segments of the series.
Claim 3
Dependent Claim
18 words
The method as in claim 1, in which said geometrical model has the form R = 1262.79 f²/r.
Claim 4
Dependent Claim
23 words
The method as in claim 1, in which said linearized model has the form: R = ≅R(f o )+[dR(f o )/df](f-f o ).
Claim 5
Dependent Claim
21 words
The method as in claim 1, in which said initialized model has the form R = β₁(s R f)+β₂R max .
Claim 6
Dependent Claim
41 words
The method as in claim 2, in which said model is initialized by (i) setting R equal to R max (desired roughness) and solving for the initial feed, and (ii) placing the initialized feed in the linearized model to complete initialization.
Claim 7
Dependent Claim
17 words
The method as in claim 5, in which the initial feed is .0281406 √ rR max .
Claim 8
Dependent Claim
99 words
The method as in claim 5, in which linearizing and initializing are carried out by (i) linearizing the static mathematical model using Taylor series expansion techniques, and (ii) substituting the initialized feed in such linearized expression to provide a mathematical model of the form R = β₁s R f+β₂R max where R is the measured or updated roughness, β₁ and β₂ are estimated coefficients, s R is a scale factor chosen to make the term s R f in the expression have the same order of magnitude as R, f is feed, and R max is the desired roughness.
Claim 9
Dependent Claim
16 words
The method as in claim 7, in which s = 35.53804 √ R max /r .
Claim 10
Dependent Claim
62 words
The method as in claim 1, in which computerized estimation of step (c) is carried out by (i) converting said linearized model to vector-matrix notation, (ii) inserting measured roughness and feed values taken from the last machined workpiece or segment into such vector model, and (iii) recursively estimating β₁ and β₂ by sequential regression analysis to provide an updated solution for feed.
Claim 11
Dependent Claim
57 words
The method as in claim 9, in which the converted form of the linearized mathematical model is R n = ϑ T x n , where R n is measured roughness, ϑ is the parameter vector of the coefficients β₁ and β₂, and x n is the computed vector of s R f and R max .
Claim 12
Dependent Claim
81 words
The method as in claim 9, in which recursive estimation is carried out, after inserting measured roughness and feed values to constitute the vector x n , by using sequential regression analysis to update the least squares estimates of ϑ in the equation: ϑ n = ϑ n-1 +k n ε n , where ϑ n-1 is computed from the immediately preceding workpiece or segment, k n is an estimation gain vector, and ε n is a current prediction error vector.
Claim 13
Dependent Claim
72 words
The method as in claim 11, in which k n is equal to P n-1 x n /(x T n P n-1 x n +λ n ), where P n-1 is the covariance matrix of regression errors known from the previous workpiece, x n is the computed vector of s R f from the previous workpiece and R max , and λ n is the discount or forgetting factor derived from theory.
Claim 14
Dependent Claim
42 words
The method as in claim 9, in which ε n = R n -R̂ n , where R n is the measured roughness and R̂ n is the predicted roughness given by R̂ n = ϑ T n -1 x n .
Claim 15
Dependent Claim
53 words
The method as in claim 12, in which the covariance matrix of parameter estimation errors can be updated by use of the following equation: P n = (I-k T n x n )P n-1 /λ n , where I is the identity matrix, and the other terms are as defined in claim 12.
Claim 16
Dependent Claim
78 words
The method as in claim 11, in which recursive estimation can be improved by modifying k n to have a square root algorithm taking the form k n = g n /(|v n |²+λ n ), where g n is S n-1 v n , v n is s T n -1 x n , λ n is a theoretically derived discount factor, and S n is a matrix which is the square root of P n .
Claim 17
Dependent Claim
26 words
The method as in claim 15, in which λ n is chosen to keep Σ n equal to a desired experimentally derived value in the algorithm:
Claim 18
Dependent Claim
33 words
The method as in claim 15, in which S n = {S n-1 -g n v T n /[|v n |²+λ n +1√ λ n (|v n |²+λ n )]}/√ λ n .
Claim 19
Dependent Claim
35 words
The method as in claim 4, in which β₁ is monitored as a predictor of tool failure so that when β₁ equals zero or becomes negative, a message is transmitted to replace the cutting tool.
Claim 20
Dependent Claim
38 words
The method as in claim 10, in which the accuracy of estimating ϑ is improved by incorporating a forgetting factor λ n chosen to minimize and keep constant the discounted sum of squared model errors Σ n .
Claim 21
Dependent Claim
30 words
The method as in claim 19, in which λ n = √ |v n |²+c n -c n , where c n = (ε n ²/Σ o +|v n |²-1)/²
Claim 22
Dependent Claim
113 words
The method as in claim 1, in which said largest allowable feed f is derived by using the largest recursively estimated feed which meets all of the following constraints: (i) resides between user determined minimum and maximum feeds, (ii) provides a feed which does not cause the feed rate to exceed a user determined maximum feed rate, (iii) provides a slenderness ratio which is between user determined minimum and maximum values for such slenderness ratio, (iv) provides that the cutting speed v satisfy the relationship v ≧ c/f, where c is a built-up edge threshold value, and (v) constrains the feed to be within a prescribed distance from the currently used feed value.
Claim 23
Independent Claim
160 words
A method of using artificial intelligence to obtain adaptive roughness control feed for improving the surface finish of a series of machined workpieces and/or a series of segments on a single workpiece, comprising: (a) initializing a linear computer model of surface roughness as a function of feed by (i) approximating a geometric relationship of centerline average roughness in terms of feed, (ii) estimating an initial feed and linearizing said relationship about such feed estimate, and (iii) converting such linearized model to vector matrix notation with provision for estimated coefficients; (b) interacting (i) measured feed and surface roughness values taken from an external interface with the immediately preceding workpiece or segment in said series, with (ii) estimated coefficient values for said initialized model by use of sequential regression analysis to provide an updated computer model; and (c) solving for the largest allowable feed in the updated model within selected machining parameter constraints for the next workpiece or segment in the series.
Claim 24
Independent Claim
160 words
A method of using artificial intelligence to obtain adaptive roughness control feed for improving the surface finish of a series of machined workpieces and/or a series of segments on a single workpiece, comprising: (a) initializing a linear computer model of surface roughness as a function of feed by (i) approximating a geometric relationship of centerline average roughness in terms of feed, (ii) estimating an initial feed and linearizing said relationship about such feed estimate, and (iii) converting such linearized model to vector matrix notation with provision for estimated coefficients; (b) interacting (i) measured feed and surface roughness values taken from an external interface with the immediately preceding workpiece or segment in said series, with (ii) estimated coefficient values for said initialized model by use of sequential regression analysis to provide an updated computer model; and (c) solving for the largest allowable feed in the updated model within selected machining parameter constraints for the next workpiece or segment in the series.
Claim 25
Dependent Claim
61 words
The method as in claim 22, in which said selected machining parameter constraints are equivalent to a pair of inequalities f₁ ≦ f ≦ f u , where f₁ = max(f min , d/s max , c bue /v, f p -Δ), and f u = min(f max , F max , F max /N, d/s min , f p +Δ).
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.