Building management system with voting-based fault detection and diagnostics
A building management system includes sensors configured to measure a plurality of monitored variables and fault detection and diagnosis (FDD) system configured to identify an operating state associated with the monitored variables. The FDD system includes a communications interface configured to receive samples of the monitored variables from the plurality of sensors. The FDD system includes a direction extractor configured to use locations, in a multidimensional modeling space, of a plurality of stored operating states to extract a direction from each of the stored operating states to each of the other stored operating states. The FDD system includes a fault diagnoser configured to use the extracted directions in a voting-based diagnosis to determine an operating state for each of the samples of the monitored variables.
1. A processing circuit comprising one or more processors and memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:
obtaining a sample of monitored variables that characterize an operation of building equipment;
using locations of a plurality of stored operating states of the building equipment in a multidimensional modeling space to determine directions, the directions comprising a direction from each stored operating state toward each other stored operating state;
using the directions in a voting-based diagnosis to determine an operating state for the sample of monitored variables; and
storing an association between the sample of the monitored variables and the operating state determined by the voting-based diagnosis.
2. The processing circuit of claim 1 , wherein the voting-based diagnosis comprises:
iteratively selecting each of the stored operating states;
recording, for each selected operating state, a vote for one or more of the other operating states for which a direction from the selected operating state to the other operating state is similar to a direction from the selected operating state to the sample; and
identifying the sample as belonging to the operating state with the most votes.
3. The processing circuit of claim 1 , the operations further comprising generating a plurality of models describing the plurality of stored operating states, each model defining a location of one of the stored operating states in the multidimensional modeling space.
4. The processing circuit of claim 3 , wherein the plurality of models are principal component analysis models, each principal component analysis model describing a corresponding operating state using a plurality of model parameters comprising:
a mean and standard deviation of a set of training samples associated with the corresponding operating state; and
a control limit for the corresponding operating state.
5. The processing circuit of claim 1 , the operations further comprising generating modeling parameters describing the plurality of stored operating states to scale the sample of the monitored variables to each of the stored operating states to provide scaled samples.
6. The processing circuit of claim 5 , the operations further comprising generating indices for the scaled samples using the modeling parameters describing the stored operating state to which the scaled samples are scaled.
7. The processing circuit of claim 6 , the operations further comprising:
comparing the generated indices with a control limit for the stored operating state to which the samples are scaled; and
identifying the samples as belonging to the stored operating state in response to the generated indices being within the control limit.
8. The processing circuit of claim 5 , the operations further comprising reconstructing the samples along each of the directions from one of the stored operating state to each of the other stored operating states.
9. The processing circuit of claim 8 , the operations further comprising:
comparing a sample index for each reconstructed sample with a control limit for the stored operating state to which the sample is scaled; and
identifying the sample as belonging to the stored operating state corresponding to the direction along which the sample is reconstructed in response to the sample index being within the control limit.
10. A method of detecting faults within a building management system (BMS), the method comprising:
obtaining a sample of monitored variables that characterize an operation of building equipment associated with the BMS;
using locations of a plurality of stored operating states of the building equipment in a multidimensional modeling space to determine directions, the directions comprising a direction from each stored operating state toward each other stored operating state;
using the directions in a voting-based diagnosis to determine an operating state for the sample of monitored variables; and
storing an association between the sample of the monitored variables and the operating state determined by the voting-based diagnosis.
11. The method of claim 10 , wherein the voting-based diagnosis comprises:
iteratively selecting each of the stored operating states;
recording, for each selected operating state, a vote for one or more of the other operating states for which the direction from the selected operating state to the other operating state is similar to the direction from the selected operating state to the sample; and
identifying the sample as belonging to the operating state with the most votes.
12. The method of claim 10 , further comprising:
recording a vote for one or more of the other stored operating states in response to a determination that the directions of the one or more of the other stored operating states are similar to the direction of the sample; and
recording a vote for an unknown operating state in response to a determination that none of the directions of the other stored operating states are similar to the direction of the sample.
13. The method of claim 10 , further comprising generating a plurality of models describing the plurality of stored operating states, each model defining a location of the stored operating state in the multidimensional modeling space.
14. The method of claim 10 , further comprising useing modeling parameters describing the plurality of stored operating states to scale the sample to each of the stored operating states.
15. The method of claim 14 , further comprising generating a sample index for each scaled sample using the modeling parameters describing the stored operating state to which the sample is scaled.
16. The method of claim 15 , further comprising:
comparing the generated sample index with a control limit for the stored operating state to which the sample is scaled; and
identifying the sample as belonging to the stored operating state to which the sample is scaled in response to the generated sample index being within the control limit.
17. The method of claim 16 , further comprising:
scaling the sample to each of the plurality of stored operating states to provide scaled samples;
determining whether each of the scaled samples belongs to the stored operating state to which the sample is scaled; and
performing the voting-based fault diagnosis in response to failing to identify the sample as belonging to any of the stored operating states.
18. The method of claim 14 , further comprising reconstructing each scaled sample along the directions from a stored operating state to each of the other stored operating states.
19. The method of claim 18 , further comprising generating a sample index for each reconstructed sample using the modeling parameters describing the stored operating state to which the sample is scaled.
20. The method of claim 19 , further comprising:
comparing the sample index for each reconstructed sample with a control limit for the stored operating state to which the sample is scaled; and
identifying the sample as belonging to the stored operating state corresponding to the direction along which the sample is reconstructed in response to the sample index being within the control limit.