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

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 F of mid term features, 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 PC of principal components, 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.

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 labeler, 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 F of mid term features, 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 PC of principal components, 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.

An apparatus for controlling an industrial CNC machine tool to cease shaping a workpiece, based on shaping tool condition, comprising:

• a. a recorder, configured for recording human audible sound while an industrial CNC machine tool is shaping a workpiece;
• b. coupled to the recorder, a time-wise audio signal slicer, configured to generate a plurality of relatively short term duration slices of the industrial recorded sound;
• c. coupled to the audio signal slicer, a noise reducer, configured to reduce noise in said plurality of short term duration slices of industrial recorded sound to generate a plurality of industrial noise reduced slices of sound;
• d. coupled to the noise reducer, a feature extractor, the feature extractor comprising:
• i. a short term analyzer, configured to, for each industrial 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 an industrial short term slice of feature values;
• ii. coupled to the short term analyzer, a mid term generator, configured to:
• A. associate a plurality of industrial short term slices of feature values together to form an industrial mid term duration frame of sound related information;
• B. extract from said industrial mid term duration frame of sound related information, industrial values of a plurality F of mid term features, including at least one time domain mid term feature and at least one frequency domain mid term feature, and associate said plurality F of industrial values of mid term features together to form an industrial mid term vector of values of sound related information;
• e. coupled to said feature extractor, a select principal component generator, configured to, based on the values of the plurality of mid term features of the industrial mid term vector, generating an industrial select principal component vector of values of select principal components, the values of select principal components of said industrial select principal component vector constituting an industrial pattern of values;
• f. coupled to the select principal component generator, a classifier configured to classify the industrial select principal component vector with respect to the condition of the shaping tool at the time the sound from which the industrial principal component vector was generated, the condition of the shaping tool being selected from the group consisting of normal and other than normal, said classification being based on:
• i. a previously prepared reference principal component information set, of a plurality of labeled mid term reference vectors of values of principal components, the values of principal components of each reference vector constituting a reference pattern of values, each reference vector also having a tool condition label associated therewith;
• and
• ii. a previously generated set of rules that match each pattern of values of a reference vector of values of principal components to a single tool condition label;
• and
• g. coupled to the classifier and the CNC machine, a shaping operation off switch, the switch configured to cease material shaping by the CNC machine if the classifier classifies an industrial principal component vector with any tool condition label other than normal.