- 上一篇：复合材料高强度混凝土的试验研究桥梁英文文献和中文翻译
- 下一篇：齿轮箱外壳综合设置和夹具规划系统英文文献和中文翻译
From thesetwo figures, it can be seen that supply air temperature increases at the beginning of fault andreaches a constant level in 10min whether it is a single fault in ST or a combined fault due toST+OD, ST+SFR or ST+SF combinations. From these results it can be reasoned that faultsdue to OD, SFR and SF elements have very limited impact when combined with the single faultin ST.Figure 4(c) shows the heater power consumption caused by single fault in ST sensor and inoutdoor damper. In contrast Figure 4(d) shows the heater power under the influence ofcombined faults in ST+OD, ST+SFR or ST+SF elements. A comparison of heater power intwo cases shows that single fault power consumption tends to be added causing higher heaterpower consumption in multiple fault operation.Similarly the supply fan pressure responses under single and multiple faults are depicted inFigures 4(e) and 4(f), respectively. These two responses are identical meaning a fault in supplyair temperature does not have any influence on the supply fan output Therefore, combined faultin ST and SF gives the same output response as the single SF fault response.Figure 4(g) shows flow rate changes caused by single fault in SFR and combined fault due toST+SFR. The former generates higher airflow rate than the latter. Figure 4(h) shows theTable V. Fault type classification.Type Variable DescriptionSingle fault OD Outdoor damperIT Indoor temperature sensorSFR Supply fan flow rateST Supply temperature sensorSF Supply fanMultiple fault ST+OD Multiple occurrenceST+SFR Multiple occurrenceST+SF Multiple occurrence output of VAV damper due to the single faults in ST sensor and SFR sensors. When supplytemperature fault occurs, the output value reduces so that VAV damper shows 40% openposition. When single fault due to SFR sensor occurs, damper opening increases progressivelyso that VAV damper opens to 50% of its full open position. These results make it possible topide the combined faults into the following three cases.First, the measurement from a single faulty element is very nearly the same as that of themeasurement from two faulty elements combined (case 1). Second, when a faulty outputmeasurement in one element is combined with another faulty measurement from a secondelement, the final measurement value is higher than single fault output (case 2). Third, faultoccurring at an element does not have any effect on the fault of other elements (case 3). Case 1 ischaracterized by similarity between combined fault and single fault. The combined fault inST+OD follows this type (Figures 4(a) and 4(b)). Case 2 shows that combined fault is the sumof the single fault values.
This type of fault is shown by combined faults in ST+SFR (Figures4(c) and 4(d)). Case 3 shows that fault of single element does not influence the combined fault.This type of fault is depicted as combined ST+SF fault in Figures 4(e) and 4(f).5.2. Diagnostic analysis of single and multiple-fault(s)Table VI shows the raw data, residuals, normalized values and resulting patterns of each faultboth in single and multiple fault modes. The raw data is the output value of each faulty element,and the residual value is the difference between fault state and normal state. Normalized value isthe residual value of each element pided by maximum residual value, using the normalizedEquation (6). By using these normalized values the fault patterns were created. The normalizedresidual magnitudes greater than 0.5 were set equal to1, and those with magnitudes less than 0.5were assigned a value of 0.5.2.1. Single fault diagnosis. A number of studies have shown the use of residual patterns indetecting single mode faults. Since, each fault gives a distinct residual pattern, these patterns are used to identify the faults. We have also analyzed the single mode fault patterns obtained in ourexperiments. These patterns are shown in Table VII. It is apparent that the residual patterns areunique to each fault category. These patterns were used as inputs to train a neural network. Thetrained neural network was tested to check its validity in identifying the faults. These results areshown as neural network outputs in Table VII.5.2.2. Multiple fault diagnosis. The residual results presented in Table VI also show somestartling differences compared to single fault residuals. For example, the heater power residualin combined ST+OD fault mode is about 35% greater than the single fault ST residual. On theother hand, in the combined fault involving ST+SF and single fault in SF the magnitude of SFresidual remains the same. When normalization is done on the basis of these residuals, however, we note that in the caseof ST+SFR, SFR has additional normalization value of 1 as compared to single fault residualin ST sensor. Similarly, in the case of ST+SF, SF has additional normalization value 1compared to the normalized residuals from the ST sensor. Furthermore, we also note that if acombined fault in ST+OD occurs, it cannot be distinguished from single fault in ST sensor.This is evident by looking at Table VIII. When a neural network is trained using such residualpatterns, it cannot distinguish single fault from a multiple fault since the same input to theneural network must identify two distinct outputs. In such cases the neural network fails toidentify the fault conclusively. To deal with this type of situations, we propose the use ofresidual ratios.