Predictive maintenance of refrigeration cases
Embodiments of the present invention disclose a method, a computer program product, and a computer system for predictive maintenance of refrigeration cases. A computer collects a temperature time series for a refrigeration case and, based on the temperature time series, learns a refrigeration case signature for both non-frost and defrost cycles. The computer generates features based on the refrigeration case signature and compares the refrigeration case signature to real time, or observed, temperatures and features using a rule-based and/or machine learning framework. Based on determining that the real time data varies beyond a threshold from the refrigeration case signature, the computer identifies a failure symptom of the refrigeration case and diagnoses a root cause of the symptom or failure. In addition, the computer may activate an alarm and open a work order corresponding to the root cause of the symptom or failure.
1. A computer-implemented method of predicting maintenance for a refrigeration case, the method comprising:
measuring real time temperature data of a refrigeration case; and
identifying a failure symptom of the refrigeration case based on comparing the real time temperature data to historical temperature data that comprises a defrost temperature signature and a non-defrost temperature signature, the identifying the failure symptom further comprising:
determining an anomaly threshold based on the defrost temperature signature;
based on determining that the real time temperature data exceeds the anomaly threshold, determining a defrost anomaly score;
decomposing the non-defrost temperature signature into seasonal, trend, and random components;
determining a drift score based on the trend component, a volatility score based on the random component, and an anomaly score based on a dynamic time warp distance between a daily temperature profile and a median temperature profile corresponding to a previous one or more days;
generating a matrix having one or more rows representing one or more days of the refrigeration case operation and one or more columns representing the defrost anomaly score, the drift score, the volatility score, and the anomaly score;
computing the matrix to output a binary variable of either a one or a zero; and
wherein identifying the failure symptom is further based on the defrost anomaly score and computing the binary value of one;
based on identifying the failure symptom, identifying one or more root causes;
assigning a confidence score to the one or more root causes;
ranking the one or more root causes based on the confidence score; and
activating an alarm and opening a work order corresponding to a root cause of the one or more root causes having a highest confidence score.
2. The method of claim 1 , wherein identifying the failure symptom of the refrigeration case based on comparing the real time temperature data to the historical temperature data further comprises:
applying one or more time series aggregation functions over various time windows to the drift score, the volatility score, and the anomaly score, wherein the one or more time series aggregation functions include at least one of a mean, a median, a max, and a range; and
wherein computing the matrix to output the binary variable is further based on the one or more columns of the matrix including the one or more time series aggregation functions.
3. The method of claim 2 , wherein identifying the failure symptom of the refrigeration case based on comparing the real time temperature data to the historical temperature data further comprises:
determining a case type score based on a case type of the refrigeration case; and
wherein computing the matrix to output the binary variable is further based on the one or more columns of the matrix including the case type score.
4. The method of claim 1 , wherein the defrost temperature signature is generated by:
collecting a temperature time series corresponding to the refrigeration case, wherein the temperature time series contains one or more defrost cycles;
filtering anomalous defrost cycles from the one or more defrost cycles by removing defrost cycles in which a preceding non-defrost cycle or subsequent non-defrost cycle did not reach a set temperature; and
identifying the defrost temperature signature as a cluster of the remaining one or more defrost cycles.
5. A computer program product for predicting maintenance for a refrigeration case, the computer program product comprising:
one or more computer-readable storage media and program instructions stored on the one or more computer-readable storage media for performing a method, the method comprising:
measuring real time temperature data of a refrigeration case; and
identifying a failure symptom of the refrigeration case based on comparing the real time temperature data to historical temperature data that comprises a defrost temperature signature and a non-defrost temperature signature, the identifying the failure symptom further comprising:
determining an anomaly threshold based on the defrost temperature signature;
based on determining that the real time temperature data exceeds the anomaly threshold, determining a defrost anomaly score;
decomposing the non-defrost temperature signature into seasonal, trend, and random components;
determining a drift score based on the trend component, a volatility score based on the random component, and an anomaly score based on a dynamic time warp distance between a daily temperature profile and a median temperature profile corresponding to a previous one or more days;
generating a matrix having one or more rows representing one or more days of the refrigeration case operation and one or more columns representing the defrost anomaly score, the drift score, the volatility score, and the anomaly score;
computing the matrix to output a binary variable of either a one or a zero; and
wherein identifying the failure symptom is further based on the defrost anomaly score and computing the binary value of one;
based on identifying the failure symptom, identifying one or more root causes;
assigning a confidence score to the one or more root causes;
ranking the one or more root causes based on the confidence score; and
activating an alarm and opening a work order corresponding to a root cause of the one or more root causes having a highest confidence score.
6. The computer program product of claim 5 , wherein the identifying the failure symptom of the refrigeration case based on comparing the real time temperature data to the historical temperature data further comprises:
applying one or more time series aggregation functions over various time windows to the drift score, the volatility score, and the anomaly score, wherein the one or more time series aggregation functions include at least one of a mean, a median, a max, and a range; and
wherein the computing the matrix to output the binary variable is further based on the one or more columns of the matrix including the one or more time series aggregation functions.
7. The computer program product of claim 6 , wherein the identifying the failure symptom of the refrigeration case based on comparing the real time temperature data to the historical temperature data further comprises:
determining a case type score based on a case type of the refrigeration case; and
wherein the computing the matrix to output the binary variable is further based on the one or more columns of the matrix including the case type score.
8. The computer program product of claim 5 , wherein the defrost temperature signature is generated by:
collecting a temperature time series corresponding to the refrigeration case, wherein the temperature time series contains one or more defrost cycles;
filtering anomalous defrost cycles from the one or more defrost cycles by removing defrost cycles in which a preceding non-defrost cycle or subsequent non-defrost cycle did not reach a set temperature; and
identifying the defrost temperature signature as a cluster of the remaining one or more defrost cycles.
9. A computer system for predicting maintenance for a refrigeration case, the computer system comprising:
one or more computer processors, one or more computer-readable storage media, and program instructions stored on one or more of the computer-readable storage media for execution of a method by at least one of the one or more processors, the method comprising:
measuring real time temperature data of a refrigeration case; and
identifying a failure symptom of the refrigeration case based on comparing the real time temperature data to historical temperature data that comprises a defrost temperature signature and a non-defrost temperature signature, the identifying the failure symptom further comprising:
determining an anomaly threshold based on the defrost temperature signature;
based on determining that the real time temperature data exceeds the anomaly threshold, determining a defrost anomaly score;
decomposing the non-defrost temperature signature into seasonal, trend, and random components;
determining a drift score based on the trend component, a volatility score based on the random component, and an anomaly score based on a dynamic time warp distance between a daily temperature profile and a median temperature profile corresponding to a previous one or more days;
generating a matrix having one or more rows representing one or more days of the refrigeration case operation and one or more columns representing the defrost anomaly score, the drift score, the volatility score, and the anomaly score;
computing the matrix to output a binary variable of either a one or a zero; and
wherein identifying the failure symptom is further based on the defrost anomaly score and computing the binary value of one;
based on identifying the failure symptom, identifying one or more root causes;
assigning a confidence score to the one or more root causes;
ranking the one or more root causes based on the confidence score; and
activating an alarm and opening a work order corresponding to a root cause of the one or more root causes having a highest confidence score.
10. The computer system of claim 9 , wherein the identifying the failure symptom of the refrigeration case based on comparing the real time temperature data to the historical temperature data further comprises:
applying one or more time series aggregation functions over various time windows to the drift score, the volatility score, and the anomaly score, wherein the one or more time series aggregation functions include at least one of a mean, a median, a max, and a range; and
wherein the computing the matrix to output the binary variable is further based on the one or more columns of the matrix including the one or more time series aggregation functions.
11. The computer system of claim 9 , wherein the defrost temperature signature is generated by:
collecting a temperature time series corresponding to the refrigeration case, wherein the temperature time series contains one or more defrost cycles;
filtering anomalous defrost cycles from the one or more defrost cycles by removing defrost cycles in which a preceding non-defrost cycle or subsequent non-defrost cycle did not reach a set temperature; and
identifying the defrost temperature signature as a cluster of the remaining one or more defrost cycles.