On-site analysis system with central processor and method of analyzing

A method of generating a calibration model for predicting a value of a property of interest from data acquired on an unknown sample of material comprising:

• obtaining a preliminary model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying at least one factor which may influence the predictive ability of the preliminary model for the property of interest;
• determining the at least one factor which influences the predictive ability of the preliminary model outside a limit of defined precision;
• and
• revising the preliminary model to compensate for variation in the at least one factor which influences the property of interest to generate the calibration model, the models predicting the value of the property of interest within the limits of defined precision.

A method of updating a calibration model to compensate for a new influential factor for predicting a property of interest, comprising:

• obtaining a calibration model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying a new factor which may influence the predictive ability of the generated calibration model for the property of interest using a validation set spanning at least a portion of a range of the new factor;
• calculating the RMSEP of the validation set to determine whether the new factor is influential;
• determining whether the new influential factor causes the predicted ability of the generated calibration model to fall outside a limit of precision defined in terms of RMSEP;
• and
• updating the generated calibration model to compensate for variance in the new influential factor to generate a revised calibration model having a predictive ability within the limit of precision.

A method of updating a calibration model to compensate for a modified range of an existing factor for predicting a property of interest, comprising:

• obtaining a calibration model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying the modified range of the existing factor which may influence the predictive ability of the generated calibration model for the property of interest using a validation set spanning at least a portion of the modified range;
• calculating the RMSEP of the validation set to determine whether the new factor is influential;
• determining whether the modified range causes the predicted ability of the generated calibration model to fall outside a limit of precision defined in terms of RMSEP;
• and
• updating the generated calibration model to compensate for variance in the new influential factor to generate a revised calibration model having a predictive ability within the limit of precision.

A method of updating a calibration model to compensate for a new influential factor for predicting a property of interest, comprising:

• obtaining a calibration model for predicting the property of interest, the model developed from a first training set using at least one instrument;
• identifying a new factor which may influence the predictive ability of the generated calibration model for the property of interest using a second training set spanning at least a portion of a range of the new factor;
• calculating the RMSECV of the validation set to determine whether the new factor is influential;
• determining whether the new influential factor causes the predicted ability of the generated calibration model to fall outside a limit of precision defined in terms of RMSECV;
• and
• updating the generated calibration model to compensate for variance in the new influential factor to generate a revised calibration model having a predictive ability within the limit of precision.

A method of updating a calibration model to compensate for a modified range of an existing factor for predicting a property of interest, comprising:

• obtaining a calibration model for predicting the property of interest, the model developed from a first training set using at least one instrument;
• identifying the modified range of the existing factor which may influence the predictive ability of the generated calibration model for the property of interest using a second training set spanning at least a portion of the modified range;
• calculating the RMSECV of the validation set to determine whether the new factor is influential;
• determining whether the modified range causes the predicted ability of the generated calibration model to fall outside a limit of precision defined in terms of RMSECV;
• and
• updating the generated calibration model to compensate for variance in the new influential factor to generate a revised calibration model having a predictive ability within the limit of precision.

A method of defining at least one acceptable region of data wherein the data are generated from an instrument response prior to evaluation of the data by a calibration model to determine if pretreatment can compensate for an influential factor in predicting a property of interest using the calibration model, comprising:

• identifying at least one trial region containing at least one subregion of data within an entire region of data from a plurality of trial regions comprising discrete combinations of subregions;
• evaluating the calibration model for each identified trial region using a validation set spanning at least a portion of a range of the influential factor;
• calculating a RMSEP for each identified trial region;
• and
• selecting at least one acceptable region from the at least one identified trial region having the RMSEP within a limit of precision identified in terms of RMSEP.

A method of defining a refined filter using a validation set, comprising:

• identifying at least one acceptable filter for the validation set;
• selecting a set of at least one acceptable filter wherein each filter within the set has the lowest number of subregions;
• selecting a subset of the set wherein each filter within the subset has the lowest rank;
• and
• defining the refined filter as the filter within the subset which further has the lowest RMSEP calculated from the validation set.

A method of defining a refined filter using a validation set, comprising:

• identifying at least one acceptable filter for the validation set;
• selecting a set of at least one acceptable filter wherein each filter within the set has the lowest rank;
• selecting a subset of the set wherein each filter within the subset has the lowest number of subregions;
• and
• defining the refined filter as the filter within the subset which further has the lowest RMSEP calculated from the validation set.

A method of defining at least one acceptable region of data wherein the data are generated from an instrument response prior to evaluation of the data by a calibration model to determine if pretreatment can compensate for an influential factor in predicting a property of interest using the calibration model, comprising:

• identifying at least one trial region containing at least one subregion of data within an entire region of data from a plurality of trial regions comprising discrete combinations of subregions;
• evaluating the calibration model for each identified trial region using a training set spanning at least a portion of a range of the influential factor;
• calculating a RMSECV for each identified trial region;
• and
• selecting at least one acceptable region from the at least one identified trial region having the RMSECV within a limit of precision identified in terms of RMSECV.

A method of defining a refined filter using a training set, comprising:

• identifying at least one acceptable filter for the training set;
• selecting a set of at least one acceptable filter wherein each filter within the set has the lowest number of subregions;
• selecting a subset of the set wherein each filter within the subset has the lowest rank;
• and
• defining the refined filter as the filter within the subset which further has the lowest RMSECV calculated from the training set.

A method of defining a refined filter using a training set, comprising:

• identifying at least one acceptable filter for the training set;
• selecting a set of at least one acceptable filter wherein each filter within the set has the lowest rank;
• selecting a subset of the set wherein each filter within the subset has the lowest number of subregions;
• and
• defining the refined filter as the filter within the subset which further has the lowest RMSECV calculated from the training set.

A method of revising a calibration model developed from a training set for a calibration model development for predicting a value of a property of interest comprising:

• detecting at least one probable outlier in the training set;
• identifying at least one good probable outlier from a group of at least one detected probable outlier;
• and
• extending the training set using the good probable outlier to develop a revised calibration model.

A method of revising a calibration model development from a training set for calibration model development for predicting a value of a property of interest comprising:

• detecting at least one probable outlier in a validation set;
• identifying at least one good probable outlier from a group of at least one detected probable outlier;
• and
• extending the training set using the good probable outlier to develop a revised calibration model.

A method of revising a calibration model developed from a training set for use in predicting a value of a property of interest on at least one unknown sample comprising:

• detecting at least one probable outlier in at least one predicted value from measurements on the at least one unknown sample;
• identifying at least one good probable outlier from a group comprising at least one detected probable outlier;
• and
• extending the training set using the good probable outlier to develop a revised calibration model.

A method of revising a calibration model developed from a training set for a calibration model development for predicting a value of a property of interest comprising:

• detecting at least one probable outlier in the training set;
• identifying at least one good probable outlier from a group of at least one detected probably outlier;
• and
• improving the training set by replacing one or more observations in the training set using the good probable outlier to develop a revised calibration model.

A method of revising a calibration model developed from a training set for calibration model development for predicting a value of a property of interest comprising:

• detecting at least one probable outlier in a validation set;
• identifying at least one good probable outlier from a group of at least one detected probable outlier;
• and
• improving the training set by replacing one or more observations in the training set using the good probable outlier to develop a revised calibration model.

A method of revising a calibration model developed from a training set for use in predicting a value of a property of interest on at least one unknown sample comprising:

• detecting at least one probable outlier in at least one predicted value from measurements on the at least one unknown sample;
• identifying at least one good probable outlier from a group comprising at least one detected probable outlier;
• and
• improving the training set by replacing one or more observations in the training set using the good probable outlier to develop a revised calibration model.

A method of evaluating an instrument for acceptability as a data acquisition device in combination with a calibration model such that a limit of precision is met by the instrument in combination with the calibration model in generating a result, comprising:

• generating a validation set with the instrument in combination with the calibration model;
• computing a RMSEP value from the validation set;
• and
• accepting the instrument if the RMSEP value is within the limit of precision.

A method of evaluating a component of an instrument for acceptability as a data acquisition device in combination with a calibration model such that a limit of precision is met by the instrument in combination with the calibration model in generating a result, comprising:

• generating a validation set with the component of the instrument in combination with the calibration model;
• computing a RMSEP value from the validation set;
• and
• accepting the component of the instrument if the RMSEP value is within the limit of precision.

A program product, comprising:

• a program configured to define at least one acceptable region of data, wherein the data are generated from an instrument response prior to evaluation of the data by a calibration model to determine if pretreatment can compensate for an influential factor in predicting a property of interest using the calibration model, by identifying at least one trial region containing at least one subregion of data within an entire region of data from a plurality of trial regions comprising discrete combinations of subregions;
• evaluating the calibration model for each identified trial region using a validation set spanning at least a portion of a range of the influential factor;
• calculating a RMSEP for each identified trial region;
• and selecting at least one acceptable region from the at least one identified trial region having the RMSEP within a limit of precision identified in terms of RMSEP, wherein the at least one acceptable region is selected based on a comparative evaluation of RMSEP, number of subregions, and rank of the calibration model, wherein lower values for each of RMSEP, number of regions and rank are preferred;
• and
• a signal bearing medium bearing the program.

A product program comprising:

• a program configured to generate a calibration model for predicting a value of a property of interest from data acquired on an unknown sample of material by obtaining a preliminary model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying at least one factor which may influence the predictive ability of the preliminary model for the property of interest;
• determining the at least one factor which influences the predictive ability of the preliminary model outside a limit of defined precision;
• and revising the preliminary model to compensate for variation in the at least one factor which influences the property of interest to generate the calibration model, the models predicting the value of the property of interest within the limits of defined precision;
• and
• a signal bearing medium bearing the program.

A program product comprising:

• a program configured to evaluate a component of an instrument for acceptability as a data acquisition device in combination with a calibration model such that a limit of precision is met by the instrument in combination with the calibration model in generating a result, by generating a validation set with the component of the instrument in combination with the calibration model;
• computing a RMSEP value from the validation set;
• and accepting the component of the instrument if the RMSEP value is within the limit of precision;
• and
• a signal bearing medium bearing the program.

An apparatus, comprising:

• at least one microprocessor;
• and
• a program configured to execute on the at least one microprocessor to define at least one acceptable region of data, wherein the data are generated from an instrument response prior to evaluation of the data by a calibration model to determine if pretreatment can compensate for an influential factor in predicting a property of interest using the calibration model, by identifying at least one trial region containing at least one subregion of data within an entire region of data from a plurality of trial regions comprising discrete combinations of subregions;
• evaluating the calibration model for each identified trial region using a validation set spanning at least a portion of a range of the influential factor;
• calculating a RMSEP for each identified trial region;
• and selecting at least one acceptable region from the at least one identified trial region having the RMSEP within a limit of precision identified in terms of RMSEP, wherein the at least one acceptable region is selected based on a comparative evaluation of RMSEP, number of subregions, and rank of the calibration model, wherein lower values for each of RMSEP, number of regions and rank are preferred.

An apparatus, comprising:

• at least one microprocessor;
• and
• a program configured to execute on the at least one microprocessor to generate a calibration model for predicting a value of a property of interest from data acquired on an unknown sample of material by obtaining a preliminary model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying at least one factor which may influence the predictive ability of the preliminary model for the property of interest;
• determining the at least one factor which influences the predictive ability of the preliminary model outside a limit of defined precision;
• and revising the preliminary model to compensate for variation in the at least one factor which influences the property of interest to generate the calibration model, the models predicting the value of the property of interest within the limits of defined precision.

An apparatus, comprising:

• at least one microprocessor;
• and
• a program configured to execute on the at least one microprocessor to evaluate a component of an instrument for acceptability as a data acquisition device in combination with a calibration model such that a limit of precision is met by the instrument in combination with the calibration model in generating a result, by generating a validation set with the component of the instrument in combination with the calibration model;
• computing a RMSEP value from the validation set;
• and accepting the component of the instrument if the RMSEP value is within the limit of precision.

An apparatus comprising:

• a memory;
• a calibration model resident in the memory and configured to compensate for instrument variance in predicting a value of a property of interest for a material;
• and
• a program configured to process data obtained from a sample of the material by using the calibration model, wherein the calibration model is configured to compensate for instrument variance in predicting the value of the property of interest.

An apparatus comprising:

• a memory;
• a calibration model resident in the memory for predicting a value of a property of interest from data acquired on an unknown sample of material;
• and
• a program configured to generate the calibration model by obtaining a preliminary model for predicting the property of interest, the model developed from a training set using at least one instrument;
• identifying at least one factor which may influence the predictive ability of the preliminary model for the property of interest;
• determining the at least one factor which influences the predictive ability of the preliminary model outside a limit of defined precision;
• and revising the preliminary model to compensate for variance in the at least one factor which influences the property of interest to generate the calibration model, the models predicting the value of the property of interest within the limits of defined precision.

An apparatus, comprising:

• a memory;
• a calibration model resident in the memory and configured for use in evaluating data;
• and
• a program configured to define at least one trial region of data wherein the data are generated from an instrument response prior to evaluation of the data by the calibration model to determine if pretreatment can compensate for an influential factor in predicting a property of interest using the calibration model by identifying at least one trial region containing at least one subregion of data within an entire region of data from a plurality of trial regions comprising discrete combinations of subregions;
• evaluating the calibration model for each identified trial region using a validation set spanning at least a portion of a range of the influential factor;
• calculating a RMSEP for each identified trial regions;
• and selecting at least one identified trial region used for generating a RMSEP within a limit of precision defined in terms of RMSEP, wherein the at least one trial region is selected based on a comparative evaluation of RMSEP, number of subregions, and rank of the calibration model, wherein lower values for each of RMSEP, number of subregions, and rank of the calibration model, wherein lower values for each of RMSEP, number of subregions and rank are preferred.

An apparatus, comprising:

• a memory;
• a calibration model resident in the memory;
• and
• a program configured to evaluate a component of an instrument for acceptability as a data acquisition devise in combination with the calibration model such that a limit of precision is met by the instrument in combination with the calibration model in generating a result, by generating a validation set with a component of the instrument in combination with the calibration model;
• computing a RMSEP value from the validation set;
• and accepting the component of the instrument if the RMSEP value is within the limit of precision.