Checkweigher performance learning systems and related methods

A method for predicting checkweigher accuracy, said method comprising:

• electrically receiving configuration data for checkweighers, said configuration data indicating pulley diameter, belt length, object type, belt speed, and frame configuration;
• performing a fast Fourier transform (FFT) trace of load cells of the checkweighers during operations comprising normal operations and accuracy validation operations to develop noise data and accuracy data for the checkweighers, wherein the FFT trace data comprises belt noise and pulley noise;
• electronically classifying the noise data by component based on frequency;
• and
• correlating the accuracy data with magnitudes of the noise data to determine threshold magnitudes on a component-specific basis reflecting acceptably accurate readings;
• for each of the magnitude thresholds exceeded for a respective component of a respective checkweigher, electronically classifying the noise data for the respective component by comparison to known, fault conditions for the respective component to generate a predicted fault condition, including by matching the noise data associated with the exceeded magnitude threshold to a known fault condition in the historical operations data having a frequency and magnitude within a predetermined margin of the noise data associated with the exceeded magnitude threshold, wherein said components including at least a belt, a pulley, and a motor for each of the checkweighers;
• generating an alert comprising the predicted fault condition;
• receiving feedback regarding actual diagnosed conditions for the respective checkweigher;
• and
• updating the predetermined margin based on the feedback using one or more machine learning techniques.

A checkweigher performance learning system, said system comprising:

• checkweighers, each comprising at least:
• a conveyor, a load cell, a controller configured to perform a fast Fourier transform (FFT) trace of the load cell to derive noise data and perform a validation of the load cell to derive accuracy data, wherein the FFT trace data comprises belt noise and pulley noise;
• user devices;
• and
• one or more servers in electronic communication with the controller each of the checkweighers and the user devices, said one or more servers comprising historical operations data associated with known fault conditions and software instructions, which when executed, configure the one or more servers to:
• receive configuration data for the checkweighers from the user devices, said configuration data indicating pulley diameter, belt length, object type, belt speed, and frame configuration;
• receive the noise data and the accuracy data from the controllers of the checkweighers;
• classify the noise data by component based on frequency of the noise data, wherein the components include at least a belt, a pulley, and a motor for each of the checkweighers;
• correlate the accuracy data with magnitudes of the noise data for each component;
• publish the correlated data to an electronically sharable and editable spreadsheet;
• receive known fault condition information from the user devices and update associated entries of the correlated data with the known fault information;
• perform a linear regression analysis of the noise data for each component to develop a magnitude threshold for each component differentiating between the magnitudes associated with the accuracy data above a predetermined acceptable accuracy threshold and the magnitudes associated with the accuracy data below the predetermined acceptable accuracy threshold;
• receive additional noise data;
• and:
• classify the additional noise data by component based on frequency;
• correlate the accuracy data with magnitude values of the noise data on a component-specific basis to determine a magnitude threshold for each component providing acceptably accurate readings:
• for each of the magnitude thresholds exceeded for a respective component of a respective checkweigher, classify the noise data for the respective component by comparison to known, fault conditions for the respective component to generate a predicted fault condition, including by matching the noise data associated with the exceeded magnitude threshold to a known fault condition in the historical operations data having a frequency and magnitude within a predetermined margin of the noise data associated with the exceeded magnitude threshold:
• generate an alert if the additional noise data incudes a magnitude above the magnitude threshold for any of the components;
• and
• include, in the alert, an indicator of a likely fault where the magnitude of the additional noise data for a given component is within a predetermined margin of a magnitude of the correlated data associated with one of the known fault conditions for the given component;
• receive feedback regarding actual diagnosed conditions for the respective checkweigher;
• and
• update the predetermined margin based on the feedback using one or more machine learning techniques.