An unknown fault type diagnosis method based on random deep forest

Abstract

This invention relates to a method for diagnosing unknown fault types based on random deep forests. Based on sample data from historical time series of air conditioning units, multiple training sets are extracted and established to train different deep forest models, obtaining separate fault diagnosis models for each air conditioning unit with preset known fault types. Then, the method applies the detection data of the air conditioning unit at a target time, and based on the consistency judgment between different diagnostic states, it achieves the diagnosis of the air conditioning unit at the target time, thus realizing accurate diagnosis of both known and unknown faults. Compared with other methods that can diagnose unknown fault types, this method has unique advantages, solving the problem of unknown faults caused by a lack of understanding of the system's physical mechanisms and experience, as well as the inability of traditional methods to detect unknown fault types, ensuring the stability of actual operation.

Description

Technical Field

[0001] This invention relates to a method for diagnosing unknown fault types based on random deep forest, belonging to the field of air conditioning unit fault diagnosis technology. Background Technology

[0002] HVAC systems account for the largest share of a building's total energy consumption, approximately 50%-60%. Equipment failures contribute about 42% of cooling energy consumption and 26% of maintenance costs within HVAC systems. It is estimated that HVAC fault diagnosis can reduce energy consumption by 10%-40%. As a crucial subsystem of HVAC systems, air conditioning units are used to regulate indoor air to a comfortable temperature. This system consists of cooling coils, fans, and other equipment. The equipment within the air conditioning unit is interconnected to control indoor temperature, humidity, and other factors to meet people's comfort needs. Current research focuses on diagnosing known fault types; studies on diagnosing unknown fault types are relatively limited. Furthermore, known fault types are often misdiagnosed as normal states or known fault types using current methods, leading to inaccurate fault location and failure to identify the cause, potentially causing serious consequences. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method for diagnosing unknown fault types based on random deep forest. It adopts a brand-new design architecture and can diagnose unknown faults in air conditioning units, thus ensuring the stability of actual operation.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention designs an unknown fault type diagnosis method based on random deep forest, and obtains the fault diagnosis model of each air conditioning unit according to the following steps A to C; then, the diagnosis of the air conditioning unit corresponding to the target time is realized according to the following steps i to ii.

[0005] Step A. Based on the historical time points of each sample corresponding to the normal state and the preset fault states of the air conditioning unit, obtain the feature vector composed of the detection data of each specified fault feature on the air conditioning unit at each sample historical time point. Based on the feature vector at the sample historical time point and combined with the corresponding state, samples are formed to obtain the samples corresponding to each sample historical time point. Then proceed to Step B. Among them, the number of sample historical time points of the air conditioning unit corresponding to any state is greater than the preset number M of deep forest used for diagnosing unknown faults, and M is greater than 1.

[0006] Step B. Extract M independent training sets from all samples, which are balanced with each other, and each training set contains samples corresponding to the normal state and preset fault states, and then proceed to step C.

[0007] Step C. Train M deep forest models using M training sets respectively to obtain the deep forest models corresponding to each training set, which are the fault diagnosis models for each air conditioning unit.

[0008] Step i. Collect feature vectors composed of the detection data of each specified fault feature on the air conditioning unit at the target time, and use them as the target feature vectors, then proceed to step ii;

[0009] Step ii. Based on the target feature vector as input, apply the fault diagnosis model of each air conditioning unit to obtain the diagnosis status output by each air conditioning unit fault diagnosis model, and determine whether the diagnosis statuses are consistent. If they are, confirm the status of the air conditioning unit at the target time and realize the air conditioning unit diagnosis; otherwise, confirm that the air conditioning unit has an unknown fault status other than the preset fault statuses at the target time.

[0010] As a preferred technical solution of the present invention: the preset deep forest model includes multi-granularity scanning and cascaded forest connected sequentially from the input end to the output end. During the execution of multi-granularity scanning, each sampling window of a preset size is first used to sample the input feature vector. Each sampling window corresponds to a forest model composed of random forest and completely random forest. The sampled feature vectors obtained by each sampling window are sent to the corresponding forest model. The random forest and completely random forest in the forest model process the received sampled feature vectors to obtain the result feature vectors. The two result feature vectors are concatenated to obtain the high-dimensional feature vector, that is, the high-dimensional feature vectors output by each forest model are obtained and sent to the cascaded forest.

[0011] As a preferred technical solution of the present invention: the cascaded forest receives high-dimensional feature vectors from multi-granularity scanning, and the output of each layer in the cascaded forest execution is sequentially concatenated with one of the high-dimensional feature vectors of each forest model in multi-granularity scanning to form a new feature vector, and then transferred to the next layer.

[0012] As a preferred technical solution of the present invention: In step ii, after confirming that the air conditioning unit has an unknown fault state other than the preset fault states of each type at the target time, the intersection of the locations of the occurrence of each diagnostic state is taken as the location of the unknown fault state, and the intersection of the causes of the occurrence of each diagnostic state is taken as the cause of the unknown fault state.

[0013] As a preferred technical solution of the present invention: In step B, based on all samples and M numerical intervals evenly divided into [0,1], each state corresponding to the air conditioning unit is taken as the state to be analyzed, and the following steps B1 to B8 are executed to realize the allocation of each sample to each numerical interval under the state to be analyzed, that is, to obtain M training sets that are independent of each other and satisfy data balance, and each training set contains samples corresponding to the normal state and preset fault states.

[0014] Step B1. Randomly select an unassigned sample in the state to be analyzed, and define [0,1] as the first value range, then proceed to step B2;

[0015] Step B2. Generate a random number within the first value range, corresponding to the sample, and proceed to step B3;

[0016] Step B3. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, return to step B2; otherwise, assign the sample to the specified numerical range and proceed to step B4; where N represents the number of samples in the state to be analyzed. Indicates rounding down;

[0017] Step B4. Determine whether the number of unassigned samples in the state to be analyzed is greater than 1. If the result is yes, return to step B1; otherwise, proceed to step B5.

[0018] Step B5. Determine if the number of unassigned samples in the state to be analyzed is equal to 0. If yes, end the assignment of samples in the state to be analyzed; otherwise, proceed to step B6.

[0019] Step B6. Randomly select an unassigned sample from the state to be analyzed, and define [0,1] as the second value range, then proceed to step B7;

[0020] Step B7. Generate a random number within the second value range that corresponds to the sample, and proceed to step B8;

[0021] Step B8. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, return to step B7; otherwise, assign the sample to the specified range and return to step B5; where, This indicates rounding up to the nearest integer.

[0022] As a preferred embodiment of the present invention: in step B3, based on M numerical intervals, it is determined that the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number corresponding to the sample belongs reaches a certain threshold. When the value range is removed from the first value range, the first value range is updated, and the process returns to step B2.

[0023] In step B8, based on M numerical intervals, it is determined that the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number corresponding to the sample belongs reaches a certain threshold. When the value range is removed from the second value range, the second value range is updated, and the process returns to step B7.

[0024] As a preferred embodiment of the present invention: the fault characteristics specified in step A include airflow. Pressure drop of the blower fan Supply air temperature T a,sup air supply flow rate Fan power Q sf Mixed air temperature T a,mix Cooling coil valve opening θ vlv,cc Return air temperature T a,rn Return air flow

[0025] The unknown fault type diagnosis method based on random deep forest described in this invention has the following technical advantages compared with the prior art:

[0026] (1) The unknown fault type diagnosis method based on random deep forest designed in this invention extracts and establishes multiple training sets based on the sample data of the corresponding historical time series of air conditioning units, and uses them to train each deep forest model to obtain each air conditioning unit fault diagnosis model for the detection of preset known fault types. Then, it applies the detection data of the air conditioning unit at the corresponding target time. Based on the judgment of the consistency between the diagnostic states, the diagnosis of the air conditioning unit at the corresponding target time is realized, that is, the accurate diagnosis of known faults and unknown faults is realized. Compared with other methods that can diagnose unknown fault types, it has unique advantages, solves the problem of unknown faults caused by people's lack of experience in the physical mechanism of the system, and the situation where traditional methods cannot detect unknown fault types, thus ensuring the stability of actual work. Attached Figure Description

[0027] Figure 1 This is a schematic diagram illustrating the process of designing an unknown fault type diagnosis method based on random deep forest in this invention;

[0028] Figure 2 This is a schematic diagram of the structure of multi-granularity scanning and cascaded forest in the deep forest of this invention. Detailed Implementation

[0029] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0030] The present invention presents a method for diagnosing unknown fault types based on random deep forests. In practical applications, such as... Figure 1 As shown, the specific design follows steps A to C to obtain the fault diagnosis model for each air conditioning unit.

[0031] Step A. Based on the historical time points of each sample corresponding to the normal state and the preset fault states of the air conditioning unit, obtain the feature vector composed of the detection data of each specified fault feature on the air conditioning unit at each sample historical time point. Based on the feature vector at the sample historical time point and combined with the corresponding state to form a sample, obtain the sample corresponding to each sample historical time point, and then proceed to Step B; wherein, the number of sample historical time points corresponding to any state of the air conditioning unit is greater than the preset number M of deep forest used for diagnosing unknown faults, and M is greater than 1.

[0032] In practical applications, the readings of various sensors on the main equipment of the air conditioning unit, such as the cooling coils and blower fans, including the flow rate of chilled water, are used. chilled water supply temperature T chw,sup The return water temperature T of chilled water chw,rn Cooling coil valve opening θ vlv,cc airflow Mixed airflow air supply flow Mixed air temperature T a,mix Supply air temperature T a,sup Pressure drop of the air supply fan And the fan power Q sf ; and certain equipment parameters, including the pressure drop ΔP of the air supply fan. sf,des The set value T of the supply air temperature sup,spt Design value of air supply volume In the embodiment application regarding step A, the selected fault characteristic design uses ten fault characteristics, specifically airflow. Pressure drop of the blower fan Supply air temperature T a,sup air supply flow rate Fan power Q sf Mixed air temperature T a,mix Cooling coil valve opening θ vlv,cc Return air temperature T a,rn Return air flow

[0033] Step B. Extract M independent training sets from all samples, which are data-balanced, and each training set contains samples corresponding to the normal state and preset fault states, and then proceed to step C.

[0034] In practical applications, in step B above, based on all samples and M numerical intervals evenly divided into [0,1], each state corresponding to the air conditioning unit is taken as the state to be analyzed, and the following steps B1 to B8 are executed to realize the allocation of each sample to each numerical interval under the state to be analyzed, that is, to obtain M training sets that are independent of each other and satisfy data balance, and each training set contains samples of the corresponding normal state and preset fault states.

[0035] Step B1. Randomly select an unassigned sample in the state to be analyzed, and define [0,1] as the first value range, then proceed to step B2.

[0036] Step B2. Generate a random number within the first value range that corresponds to the sample, and proceed to step B3.

[0037] Step B3. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, then remove the corresponding value interval from the first value range, update the first value range, and return to step B2; otherwise, assign the sample to the corresponding value interval, and then proceed to step B4; where N represents the number of samples in the state to be analyzed. This indicates rounding down to the nearest integer.

[0038] Step B4. Determine whether the number of unassigned samples in the state to be analyzed is greater than 1. If the result is yes, return to step B1; otherwise, proceed to step B5.

[0039] Step B5. Determine if the number of unassigned samples in the state to be analyzed is equal to 0. If yes, end the assignment of each sample in the state to be analyzed; otherwise, proceed to step B6.

[0040] Step B6. Randomly select an unassigned sample in the state to be analyzed, and define [0,1] as the second value range, then proceed to step B7.

[0041] Step B7. Generate a random number within the second value range that corresponds to the sample, and proceed to step B8.

[0042] Step B8. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, then remove the value interval from the second value range, update the second value range, and return to step B7; otherwise, assign the sample to the value interval, and then return to step B5; where, This indicates rounding up to the nearest integer.

[0043] Step C. Train M deep forest models using M training sets respectively, and obtain the deep forest model corresponding to each training set, which is the fault diagnosis model of each air conditioning unit.

[0044] In practical applications, the preset deep forest model described here includes multi-granularity scans and cascaded forests connected sequentially from the input end to the output end. Granularity scans are the first part of the deep forest, while multi-granularity scans are designed to enhance the characteristics of the data.

[0045] In the multi-granularity scanning process, the input feature vectors are first sampled using sampling windows of different preset sizes. Each sampling window corresponds to a forest model composed of random forest and fully random forest. The sampled feature vectors obtained by each sampling window are sent to the corresponding forest model. The random forest and fully random forest in the forest model process the received sampled feature vectors to obtain the result feature vectors. The two result feature vectors are then concatenated to obtain the high-dimensional feature vector, which is the output of each forest model. This high-dimensional feature vector is then sent to the cascaded forest.

[0046] This patent uses both 2D and 3D windows. The multi-granularity scanning process using the 3D window is as follows: Figure 2 As shown, considering the 10 fault features involved in the embodiment, i.e., considering that the fault features are 10-dimensional, the input sequence of the deep forest is 10-dimensional (m=10) because the scanning window is 3-dimensional (q=3). According to the formula below, when the moving window slides in one dimension (k=1), 8 instances can be generated (S=8), because the scanning window is 3-dimensional, so each instance is 3-dimensional.

[0047]

[0048] Then, as Figure 2 As shown, the eight generated instances are input into Random Forest B1 and Fully Random Forest B2. Since the output is divided into 14 classes, each instance input into a Random Forest will produce 14 outputs. Therefore, the eight instances input into B1 produce a total of 8 × 14 = 112 outputs. Similarly, inputting these eight instances into Fully Random Forest B2 will also produce 112 outputs. In other words, through B1 and B2, the original data is enhanced into a 224-dimensional feature vector. Similarly, using a 2-dimensional scanning window, a 252-dimensional feature vector can be obtained. Then, cascaded forests are needed to iterate through the data features layer by layer.

[0049] like Figure 2As shown, the cascaded forest receives high-dimensional feature vectors from multi-granularity scans. During execution, the output of each layer is concatenated with one of the high-dimensional feature vectors from each forest model in the multi-granularity scans, forming a new feature vector, which is then transferred to the next layer. For example, the 252-dimensional feature vector obtained from forest A in the multi-granularity scan is input into the cascaded forest. This 252-dimensional vector serves as the input to four forests. Since the data is divided into 14 classes, each forest outputs 14. Thus, the four forests generate a total of 4 × 14 = 56 outputs to layer 1. Then, the 252-dimensional feature vector obtained from forest A in the multi-granularity scan is directly concatenated to this 56-dimensional feature vector. Finally, a 308-dimensional feature vector is obtained, which is... Figure 2 Layer 1 in the diagram. Similarly, we can obtain layer 2, ..., layer N.

[0050] Furthermore, since decision trees and random forests can determine the entropy increase between layers, cascaded forests can adaptively determine the number of layers. When increasing the number of layers no longer significantly improves the classification performance, the increase in layers will stop, and the classification results of each random forest will be comprehensively analyzed to propose the final classification result.

[0051] Based on the fault diagnosis models of each air conditioning unit obtained above, the diagnosis of the air conditioning unit corresponding to the target time is then realized by following steps i to ii.

[0052] Step i. Collect feature vectors composed of the detection data of each specified fault feature on the air conditioning unit at the target time, and use them as the target feature vectors, and then proceed to step ii.

[0053] Step ii. Based on the target feature vector as input, apply the fault diagnosis model of each air conditioning unit to obtain the diagnosis status output by each air conditioning unit fault diagnosis model, and determine whether the diagnosis statuses are consistent. If they are, confirm the status of the air conditioning unit at the target time and realize the air conditioning unit diagnosis; otherwise, confirm that the air conditioning unit has an unknown fault status other than the preset fault statuses at the target time. Then, combine experience and physical mechanisms to realize the diagnosis of the unknown fault. The intersection of the locations of the occurrence of each diagnosis status is taken as the location of the unknown fault status, and the intersection of the causes of the occurrence of each diagnosis status is taken as the cause of the unknown fault status.

[0054] In practical applications, for example, performing the above step ii based on the three generated air conditioning unit fault diagnosis models, these three air conditioning unit fault diagnosis models may be able to diagnose unknown faults. 14Data showing that the return air fan is stuck at a fixed speed produces different classification results. For example, when fault f7 (complete failure of the return air fan) occurs, air conditioning unit fault diagnosis model 1 determines that the probability of f7 occurring is 0.8, which is the highest among all categories. Therefore, this known type of fault is classified as f7. Similarly, air conditioning unit fault diagnosis model 2, when diagnosing this known fault, can also determine that the known fault type is f7 with a probability of 0.85. Likewise, air conditioning unit fault diagnosis model 3 also determines that the fault is known fault type f7. Thus, it can be seen that different air conditioning unit fault diagnosis models can give a unified conclusion when facing known faults. However, in the case of unknown fault f... 14 When the fault occurs, because fault diagnosis models 1, 2, and 3 of the air conditioning unit have not learned the unknown fault f... 14 Because of the limited feature information, when classifying based on unknown fault type data, the probabilities of each category are not significantly different. For example, the probability of air conditioner unit fault diagnosis model 1 classifying it as f7 is 0.35, the highest among all probabilities. Therefore, air conditioner unit fault diagnosis model 1 classifies this unknown fault as f7. Similarly, air conditioner unit fault diagnosis model 2, when faced with f... 14 When the probability of judging f8 is 0.4, it is the highest. Therefore, the result of the air conditioning unit fault diagnosis model 2 is f8, while the air conditioning unit fault diagnosis model 3 judges f8 as the highest probability. 14 When the probability of determining f9 is 0.5, it is the highest. Therefore, the result of fault diagnosis model 2 for air conditioning units is f9. Thus, these three fault diagnosis models for air conditioning units are effective when facing unknown fault types f. 14 When the given judgments are inconsistent, it can be assumed that an unknown fault has occurred. The intersection of the locations of each diagnostic state is then used as the location of the unknown fault state, and the intersection of the causes of each diagnostic state is used as the cause of the unknown fault state.

[0055] The aforementioned technical solution, based on a random deep forest-based unknown fault type diagnosis method, extracts and establishes multiple training sets from historical time-series sample data of air conditioning units. These sets are used to train various deep forest models, resulting in individual fault diagnosis models for each air conditioning unit, each designed to detect pre-defined known fault types. The method then applies these models to the detection data of the air conditioning units at a target time. By judging the consistency between different diagnostic states, the system achieves accurate diagnosis of both known and unknown faults at that time. Compared to other methods, this approach offers unique advantages, addressing the issue of unknown faults caused by a lack of understanding of the system's physical mechanisms and experience, as well as the inability of traditional methods to detect unknown fault types, thus ensuring the stability of practical operation.

[0056] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A method for diagnosing unknown fault types based on random deep forest, characterized in that: Follow steps A to C to obtain the fault diagnosis model for each air conditioning unit; then follow steps i to ii to realize the diagnosis of the air conditioning unit corresponding to the target time. Step A. Based on the historical time points of each sample corresponding to the normal state and the preset fault states of the air conditioning unit, obtain the feature vector composed of the detection data of each specified fault feature on the air conditioning unit at each sample historical time point. Based on the feature vector at the sample historical time point and combined with the corresponding state, samples are formed to obtain the samples corresponding to each sample historical time point. Then proceed to Step B. Among them, the number of sample historical time points of the air conditioning unit corresponding to any state is greater than the preset number M of deep forest used for diagnosing unknown faults, and M is greater than 1. Step B. Extract M independent training sets from all samples, which are balanced with each other, and each training set contains samples corresponding to the normal state and preset fault states, and then proceed to step C. Step C. Train M deep forest models using M training sets respectively to obtain the deep forest models corresponding to each training set, which are the fault diagnosis models for each air conditioning unit. The deep forest model includes multi-granularity scanning and cascaded forests connected sequentially from the input to the output. During multi-granularity scanning, sampling windows of different preset sizes are first used to sample the input feature vectors. Each sampling window corresponds to a forest model composed of random forest and fully random forest. The sampled feature vectors obtained by each sampling window are sent to the corresponding forest model. The random forest and fully random forest in the forest model process the received sampled feature vectors to obtain the result feature vectors. The two result feature vectors are concatenated to obtain the high-dimensional feature vector, that is, the high-dimensional feature vectors output by each forest model are obtained and sent to the cascaded forest. Step i. Collect feature vectors composed of the detection data of each specified fault feature on the air conditioning unit at the target time, and use them as the target feature vectors, then proceed to step ii; Step ii. Based on the target feature vector as input, apply the fault diagnosis model of each air conditioning unit to obtain the diagnosis status output by each air conditioning unit fault diagnosis model, and determine whether the diagnosis statuses are consistent. If they are, confirm the status of the air conditioning unit at the target time to realize the diagnosis of known fault types of the air conditioning unit; otherwise, confirm that an unknown fault status other than the preset fault statuses of each type has occurred at the target time of the air conditioning unit. Use the intersection of the occurrence locations of each diagnosis status as the location of the unknown fault status, and use the intersection of the occurrence causes of each diagnosis status as the occurrence cause of the unknown fault status to realize the detection of unknown fault types.

2. The method for diagnosing unknown fault types based on random deep forest according to claim 1, characterized in that: The cascaded forest receives high-dimensional feature vectors from multi-granularity scans. During the execution of the cascaded forest, the output of each layer is concatenated with one of the high-dimensional feature vectors of each forest model in the multi-granularity scans to form a new feature vector, which is then transferred to the next layer.

3. The method for diagnosing unknown fault types based on random deep forest according to claim 1, characterized in that: In step B, based on all samples and M numerical intervals evenly divided into [0,1], each state corresponding to the air conditioning unit is taken as the state to be analyzed, and the following steps B1 to B8 are executed to realize the allocation of each sample to each numerical interval under the state to be analyzed, that is, to obtain M training sets that are independent of each other and satisfy data balance, and each training set contains samples of the corresponding normal state and preset fault states. Step B1. Randomly select an unassigned sample in the state to be analyzed, and define [0,1] as the first value range, then proceed to step B2; Step B2. Generate a random number within the first value range, corresponding to the sample, and proceed to step B3; Step B3. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, return to step B2; otherwise, assign the sample to the specified numerical range and proceed to step B4; where N represents the number of samples in the state to be analyzed. Indicates rounding down; Step B4. Determine whether the number of unassigned samples in the state to be analyzed is greater than 1. If the result is yes, return to step B1; otherwise, proceed to step B5. Step B5. Determine if the number of unassigned samples in the state to be analyzed is equal to 0. If yes, end the assignment of samples in the state to be analyzed; otherwise, proceed to step B6. Step B6. Randomly select an unassigned sample from the state to be analyzed, and define [0,1] as the second value range, then proceed to step B7; Step B7. Generate a random number within the second value range that corresponds to the sample, and proceed to step B8; Step B8. Based on M numerical intervals, determine whether the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number belongs to the sample has reached the required level. If yes, return to step B7; otherwise, assign the sample to the specified range and return to step B5; where, This indicates rounding up to the nearest integer.

4. The method for diagnosing unknown fault types based on random deep forest according to claim 3, characterized in that: In step B3, based on M numerical intervals, it is determined that the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number corresponding to the sample belongs reaches a certain threshold. When the value range is removed from the first value range, the first value range is updated, and the process returns to step B2. In step B8, based on M numerical intervals, it is determined that the number of samples corresponding to the state to be analyzed in the numerical interval to which the random number corresponding to the sample belongs reaches a certain threshold. When the value range is removed from the second value range, the second value range is updated, and the process returns to step B7.

5. The method for diagnosing unknown fault types based on random deep forest according to claim 1, characterized in that: Step A specifies various fault characteristics, including airflow. Pressure drop of the blower fan Supply air temperature T a,sup air supply flow rate Fan power Q sf Mixed air temperature T a,mix Cooling coil valve opening θ vlv,cc Return air temperature T a,rn Return air flow

Images