Method and apparatus for efficient use of CNC machine shaping tool including cessation of use no later than the onset of tool deterioration by monitoring audible sound during shaping

• US 10,245,698 B2
• Filed: 05/04/2017
• Issued: 04/02/2019
• Est. Priority Date: 05/06/2016
• Status: Active Grant

• Current Assignee: Massachusetts Institute of Technology
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Jeong, Hyunsoo
• Primary Examiner(s): Lo, Kenneth M
• Assistant Examiner(s): Erdman, Chad G
• Application Number:
  US15/586,695
• Publication Number:
  US 20170320182A1
• Time in Patent Office: 698 Days
• CPC Class Codes: B23Q 17/098 by measuring noise
• CPC Class Codes: G05B 19/4065 Monitoring tool breakage, life or condition
• CPC Class Codes: G05B 2219/33002 Artificial intelligence AI, expert, knowledge, rule based system KBS
• CPC Class Codes: G05B 2219/37337 Noise, acoustic emission, sound

Abstract

A CNC machine shaping tool is efficiently used by monitoring human audible sound during shaping. A sound information set is created for a tool shaping a workpiece. Shaping sounds are recorded and sliced into short term units. A human operator assigns tool condition labels to each slice. Short term units are combined into mid term units. Noise is reduced by profiling. Mid term sound related features of time and frequency domains are extracted. Dimensionality is reduced by robust principal component analysis. The principal component set is balanced, e.g. by SMOTE. A classifier and principal components are selected. An information set of patterns of values of selected principal components for the tool is created. In an industrial setting, shaping sounds are recorded, noise reduced and select principal component vector values are compared to the tool condition labeled patterns of values in the information set to identify tool condition before deterioration.

Claims

1. A method for generating a reference information set of CNC machine tool shaping sound related information and rules for generating tool condition labels based thereon, the method comprising the steps of:

a. recording human audible sound while a CNC machine tool is shaping a workpiece, including during normal operation and during deteriorating operation;

b. generating a plurality of relatively short term duration slices of the recorded sound;

c. selecting, by a human, for each said short term duration slice of sound, a shaping tool condition label, which selection has been made by the human based on listening to the sound generated while the machine is shaping a workpiece and judging the condition of the shaping tool and selecting a tool condition label that corresponds with the judged condition, and associating the selected shaping tool condition label with the corresponding short term duration slice of sound, resulting in a labeled short term slice of sound;

d. reducing noise in said plurality of short term duration slices of recorded labeled sound to generate a plurality of noise reduced slices of labeled sound;

e. for each labeled short term slice of sound, extracting values of a plurality of short term features, at least one of which is a time domain feature and at least one of which is a frequency domain feature, resulting in a labeled short term slice of feature values;

f. associating a plurality of labeled short term slices of feature values together to form a labeled mid term duration frame of sound related information;

g. extracting from said labeled mid term duration frame of sound related information, values of a plurality of mid term features (F), including at least one time domain mid term feature and at least one frequency domain mid term feature, and associating said plurality F of mid term features together to form a labeled mid term vector of values of sound related information;

h. repeating the steps 1.a through 1.g a plurality of times, to generate a labeled feature information set of a plurality of tool condition labeled mid term vectors of values of sound related information;

i. based on the values of the plurality of mid term features of the feature information set, generating a plurality of principal components (PC), where the plurality PC is smaller than the plurality F, and then generating, corresponding to the feature information set, a principal component information set, of a plurality of labeled mid term vectors of values of principal components, the values of principal components of each vector constituting a pattern of values, each vector also having a tool condition label associated therewith, the plurality of mid term vectors of values of principal components, the pattern thereof, and tool condition labels together constituting a principal component information set;

j. if the principal component information set is imbalanced with respect to each of the tool condition labels, then balancing the principal component information set with respect to each of the tool condition labels to generate a balanced reference principal component information set; and

k. choosing a specific classifier and classifying the balanced reference principal component information set with regard to the tool condition labels to generate rules that match each pattern of values of a vector of values of principal components to a single tool condition label.

2. The method of claim 1, the step of classifying comprising selecting a specific classifier to apply to the reference principal component information set, further comprising the steps of:

a. identifying at least two subsets of principal components;

b. evaluating each of the subsets of principal components with the classifier to determine which subset of principal components works best with the classifier; and

c. designating the principal components of the subset that works best as selected principal components.

3. The method of claim 1, the step of classifying comprising selecting a classifier, the step of selecting a classifier comprising:

a. selecting a plurality of different classifiers;

b. evaluating each of the classifiers using the balanced reference principal component information set, and based on the evaluation, identifying one of the classifiers as the best.

4. The method of claim 3, the step of evaluating each classifier being selected from the group consisting of:

evaluating each classifier'"'"'s accuracy; and

evaluating, for each classifier, ROC area.

5. The method of claim 1, the step of reducing noise comprising:

a. operating the CNC machine tool when the shaping tool is not shaping material;

b. recording human audible sound while the CNC machine tool is operating and not shaping material;

c. based on the recording of sound while the machine tool is not shaping material, generating a profile of noise that relates to sounds other than shaping material; and

d. combining the noise profile with the plurality of relatively short term duration slices of the recorded sound in a manner that reduces the effect of noise that relates to sounds other than shaping material.

6. The method of claim 1, the step of balancing the principal component information set with respect to each of the tool condition labels comprising applying a SMOTE method to the principal component information set to generate a balanced principal component information set.

7. The method of claim 1, the step of generating a smaller plurality of principal components, comprising applying Robust principal component analysis to the feature information set.

8. The method of claim 2, said step of evaluating each of the subsets of principal components with the classifier to determine which subset of principal components works best with the classifier comprising applying a wrapper method.

9. The method of claim 3, the step of selecting a plurality of different classifiers comprising selecting a classifier from the group consisting of:

Naï

ve Bayes, Multi-Class Support Vector Machine (SVM);

Multi-Class Alternative Decision Tree (LADTree);

Multinomial Logistic Regression;

k-Nearest Neighbors (kNN);

Decision Tree;

Bagging (REPTree) and Artificial Neural Networks.

10. The method of claim 1, the step of selecting a shaping tool condition label comprising judging the condition of the shaping tool with respect to at least one additional operating condition in addition to normal and deteriorating operation.

11. The method of claim 1, said plurality of short term features being selected from the group consisting of:

zero-crossing rate;

entropy of energy;

energy;

first thirteen Mel-frequency Cepstral Coefficients;

spectral entropy;

spectral rolloff;

spectral centroid;

spectral spread;

spectral flux;

harmonic pitch class profiles; and

a twelve element chroma vector.

12. The method of claim 11, the mid term features being selected from, for any one of said short term features, taking any statistical treatment of the group consisting of:

mean, median, standard deviation, standard deviation by mean, maximum, and minimum.

13. The method of claim 1, further comprising steps to identify industrial tool condition, those identification steps comprising:

a. in an industrial setting, using a CNC machine of the same type as the CNC machine that was used to generate the reference information set to shape a workpiece of the same type that was shaped to generate the reference information set, and recording human audible sound while the CNC machine tool is shaping the workpiece;

b. generating a plurality of relatively short term duration slices of the recorded industrial sound;

c. reducing noise in said plurality of short term duration slices of recorded industrial sound to generate a plurality of noise reduced slices of industrial sound;

d. for each short term slice of industrial sound, extracting values of the same plurality of short term features as was used to generate said short term slice of feature values, resulting in an industrial short term slice of feature values;

e. associating a plurality of industrial short term slices of feature values together to form an industrial mid term duration frame of sound related information;

f. extracting from said industrial mid term duration frame of sound related information, values of the plurality F of mid term features, and associating said plurality F of mid term features together to form an industrial mid term vector of values of sound related information;

g. based on the values of the plurality F of mid term features of the industrial mid term vector of values of sound related information, generating an industrial principal component information vector, the values of the principal components of the industrial principal component vector constituting a pattern of values; and

h. using the rules generated by the classifier that was used to classify the reference principal component information set, applying the rules to the pattern of values of the industrial principal component vector to identify the tool condition that matches the pattern of values of the industrial principal component vector.

14. The method of claim 13, further comprising the step of testing the identified tool condition and, if the tool condition is normal, continuing to use the CNC machine to shape the workpiece, else, stop using the CNC machine to shape the workpiece.

15. The method of claim 13, the step of reducing the noise in said plurality of short term duration slices of recorded industrial sound comprising:

a. before the step of, in an industrial setting using the CNC machine to shape a workpiece;

i. operating the CNC machine tool in the industrial setting when the shaping tool is not shaping material;

ii. recording human audible sound while the CNC machine tool in the industrial setting is operating and not shaping material;

iii. based on the recording of sound while the machine tool in the industrial setting is not shaping material, generating a profile of noise that relates to sounds in the industrial setting other than shaping material; and

b. combining the noise profile with the plurality of relatively short term duration slices of the recorded industrial sound in a manner that reduces the effect of noise that relates to industrial sounds other than shaping material.

16. A method for identifying tool condition of a shaping tool in an industrial setting, using a reference information set of CNC machine tool shaping sound related information and rules for generating tool condition labels based thereon, the method comprising the steps of:

a. recording human audible sound while a CNC machine tool is shaping a workpiece, including during normal operation and during deteriorating operation;

b. generating a plurality of relatively short term duration slices of the recorded sound;

c. selecting, by a human, for each said short term duration slice of sound, a shaping tool condition label, which selection has been made by the human based on listening to the sound generated while the machine is shaping a workpiece and judging the condition of the shaping tool and selecting a tool condition label that corresponds with the judged condition, and associating the selected shaping tool condition label with the corresponding short term duration slice of sound, resulting in a labeled short term slice of sound;

d. reducing noise in said plurality of short term duration slices of recorded sound to generate a plurality of labeled noise reduced slices of sound;

e. for each labeled noise reduced short term slice of sound, extracting values of a plurality of short term features, at least one of which is a time domain feature and at least one of which is a frequency domain feature, resulting in a labeled short term slice of feature values;

f. associating a plurality of labeled short term slices of feature values together to form a labeled mid term duration frame of sound related information;

g. extracting from said labeled mid term duration frame of sound related information, values of a plurality of mid term features (F), including at least one time domain mid term feature and at least one frequency domain mid term feature, and associating said plurality F of mid term features together to form a labeled mid term vector of values of sound related information;

h. repeating the steps 1.a through 1.g a plurality of times, to generate a labeled feature information set of a plurality of tool condition labeled mid term vectors of values of sound related information;

i. based on the values of the plurality of mid term features of the feature information set, generating a plurality of principal components (PC), where the plurality PC is smaller than the plurality F, and then generating, corresponding to the feature information set, a principal component information set, of a plurality of labeled mid term vectors of values of principal components, the values of principal components of each vector constituting a pattern of values, each vector also having a tool condition label associated therewith, the plurality of mid term vectors of values of principal components, the pattern thereof, and tool condition labels together constituting a principal component information set;

j. if the principal component information set is imbalanced with respect to each of the tool condition labels, then balancing the principal component information set with respect to each of the tool condition labels to generate a balanced reference principal component information set;

k. choosing a specific classifier and classifying the reference principal component information set with regard to the tool condition labels to generate rules that match each pattern of values of a vector of values of principal components to a single tool condition label;

l. in an industrial setting, using a CNC machine of the same type as the CNC machine that was used to generate the reference information set to shape a workpiece of the same type that was shaped to generate the reference information set, and recording human audible sound while the CNC machine tool is shaping the workpiece;

m. generating a plurality of relatively short term duration slices of the recorded industrial sound;

n. reducing noise in said plurality of short term duration slices of recorded industrial sound to generate a plurality of noise reduced slices of industrial sound;

o. for each labeled short term slice of industrial sound, extracting values of the same plurality of short term features, used to generate said short term slice of feature values, resulting in an industrial short term slice of feature values;

p. associating a plurality of industrial short term slices of feature values together to form an industrial mid term duration frame of sound related information;

q. extracting from said industrial mid term duration frame of sound related information, values of the plurality F of mid term features, and associating said plurality F of mid term features together to form an industrial mid term vector of values of sound related information;

r. based on the values of the plurality F of mid term features of the industrial mid term vector of values of sound related information, generating an industrial principal component information vector, the values of the principal components of the industrial principal component vector constituting a pattern of values; and

s. using the rules generated by the classifier that was used to classify the reference principal component information set, applying the rules to the pattern of values of the industrial principal component vector to identify the tool condition that matches the pattern of values of the industrial principal component vector.

17. The method of claim 16, further comprising the step of testing the identified tool condition and, if the tool condition is normal, continuing to use the CNC machine to shape the workpiece, else, stop using the CNC machine to shape the workpiece.

18. An apparatus for generating a reference information set of CNC machine tool shaping sound related information and rules for generating tool condition labels based thereon, the method comprising:

a. a recorder, configured for recording human audible sound while a CNC machine tool is shaping a workpiece, including during normal operation and during deteriorating operation;

b. coupled to the recorder, a time-wise audio signal slicer, configured to generate a plurality of relatively short term duration slices of the recorded sound;

c. a tool condition labeller, configured to assign, to each said short term duration slice of sound, a shaping tool condition label, which shaping tool condition label has been selected by a human operator based on listening to the sound generated while the machine is shaping a workpiece and judging the condition of the shaping tool and selecting a tool condition label that corresponds with the judged condition, and assigning the selected shaping tool condition label to the corresponding short term duration slice of sound, resulting in a labeled short term slice of sound;

d. a noise reducer, configured to reduce noise in said plurality of short term duration slices of recorded sound to generate a plurality of noise reduced slices of sound;

e. coupled to the noise reducer, a feature extractor, comprising;

i. a short term analyzer, configured to, for each labeled short term slice of sound, extract values of a plurality of short term features, at least one of which is a time domain feature and at least one of which is a frequency domain feature, resulting in a labeled short term slice of feature values;

ii. coupled to the short term analyzer, a mid term generator, configured to;

A. associate a plurality of short term slices of feature values together to form a labeled mid term duration frame of sound related information;

B. extract from said labeled mid term duration frame of sound related information, values of a plurality of mid term features (F), including at least one time domain mid term feature and at least one frequency domain mid term feature, and associate said plurality F of mid term features together to form a labeled mid term vector of values of sound related information; and

C. to repeatedly form mid term duration frames of sound related information and extracting from each formed mid term duration frame of sound information, mid term features and associate each said plurality F of mid term features together to form a plurality of labeled mid term vectors of values of sound related information to generate a labeled feature information set of a plurality of tool condition labeled mid term vectors of values of sound related information;

f. coupled to said feature extractor, a principal component generator, configured to, based on the values of the plurality of mid term features of the feature information set, generate a plurality of principal components (PC), where the plurality PC is smaller than the plurality F, and then generate, corresponding to the feature information set, a principal component information set, of a plurality of labeled mid term vectors of values of principal components, the values of principal components of each vector constituting a pattern of values, each vector also having a tool condition label associated therewith, the plurality of mid term vectors of values of principal components, the pattern thereof, and tool condition labels together constituting a principal component information set;

g. coupled to the principal component generator, an information set balancer, configured to, if the principal component information set is imbalanced with respect to each of the tool condition labels, then balancing the principal component information set with respect to each of the tool condition labels to generate a balanced reference principal component information set; and

h. a classifier configured to classify the reference principal component information set with regard to the tool condition labels to generate rules that match each pattern of values of a vector of values of principal components to a single tool condition label.

19. The apparatus of claim 18, the classifier comprising a specific classifier, further comprising, coupled to the information set balancer and the classifier, a principal component selector, configured to:

a. identify at least two subsets of principal components;

b. evaluate each of the subsets of principal components with the classifier to determine which subset of principal components works best with the classifier; and

c. designating the principal components of the subset that works best as selected principal components.

20. The apparatus of claim 18, the noise reducer comprising:

a. a noise profiler configured to create a profile of human audible sound that arises while the CNC machine tool is operating and not shaping material of;

b. coupled to the noise profiler, a combiner, configured to combine the noise profile with the plurality of relatively short term duration slices of the recorded sound in a manner that reduces the effect of noise that relates to sounds other than shaping material.

21. The apparatus of claim 18, the information set balancer comprising a balancer configured to apply a SMOTE method to the principal component information set to generate a balanced principal component information set.

22. The apparatus of claim 18, the principal component generator comprising a generator configured to apply a Robust principal component analysis.

23. The apparatus of claim 19, the principal component selector comprising a wrapper method principal component selector.

24. The apparatus of claim 18, the classifier being selected from the group consisting of:

Naï

ve Bayes, Multi-Class Support Vector Machine (SVM);

Multi-Class Alternative Decision Tree (LADTree);

Multinomial Logistic Regression;

k-Nearest Neighbors (kNN);

Decision Tree;

Bagging (REPTree) and Artificial Neural Networks.