Wireless HVAC compressor diagnostics using state of the art machine learning-based signal analysis Z-freq 2D

One of the priorities in detecting a faulty car engine is through a method known as diagnostic, and it is very crucial as each diagnostic able to provide information and assessment to identify problems in the car A/C compressor. Early detection of a compressor malfunction, is a fast way to prevent any heavy maintenance of vehicles either in the short or long term. This paper introduced a new statistical method to find the faulty in the vehicle’s A/C compressor which is known as Z-freq 2D. The foundation of Z-freq 2D is involving the implementation of a Z-notch frequency domain filter. This approach was enhanced by using a special sensor that can detect two axial axes known as the Phantom Vibration Sensor specifically designed to detect and monitor the performance of the A/C compressor of the vehicle. Using the sensor, the data were recorded at numerous parameter sets of compressor speed. The analyzed data shows that Z-freq 2D coefficient is increase as the speed of the compressor increase over the duration of time. Z-freq 2D can be used to detect the malfunction and the irregularities of vibration signals, which may be indicated that the compressor is failing.


INTRODUCTION
A consistent annual checkup for the vehicle compressor especially for HEV's compressor, will help to prevent any problems rise for the driver such as car breakdown or poor driving while on the road caused by driver loss of attention when driving in hot air.Currently, there is a system to detect faulty in the vehicle engine, known as "Engine Management Light" that lights up whenever there is "something" may be wrong in the engine, indicating there is a potential issue raised from the engine however it is unable to accurately pinpoint the exact and specific faulty information regarding the issue.
Improper diagnostic of the vehicle also will cause some faults in the vehicle to be overlooked.In the long term, would cause bad effects on both the vehicle and people in the cabin.Some examples are gaseous from the HVAC leaks into the vehicle.From an average human point of view it may be nothing to worry about, from a medical and hazard point of view it is a very concerning case and could lead to death.Originally the HVAC system should provide safe and fresh air into the cabin but, a bad diagnostic to find the leak which releases harmful gases such as Carbon Dioxide (CO2) will cause numerous effects from headache, and vomiting to death (Prabhu, S., et al. 2018).
Faulty in vehicle compressors or HVAC systems, however currently are detected through the level of temperature coming out from the vehicle air-conditioning.The high hot air temperature that comes out of the A/C vault, indicates that there are faults in the vehicle's HVAC system and insufficient air cooling is the main indicator that the vehicle's air conditioning having a breakdown.According to Don't Blame The Compressor (Kimberly Schwartz.2016), stated that the two most common causes of compressor failure are the loss of lubrication and slugging.
There is numerous method to study the automotive refrigeration system for example using computational method simulation to simulate how the HVAC system are working in 3 dimension form (Joudi, K. A., et al. 2003).However, as for experimental setup, it can be done either through an actual vehicle or through a test rig which is a piece of machine that we can use for some experimental tests (Hydrotechnik UK Limited.2013) and in this case of research study, it will be an automotive refrigeration system built outside of the vehicle which still comprises of basic air conditioning system components.
The purpose of this research study, will focus on the implementation of a wireless HVAC compressor diagnostics using state-of-the-art machine learning-based signal analysis Z-freq 2D, correlating the various parameters, the speed of the compressor, the amount of refrigerant R134a, and the volume of the oil.

EXPERIMENTAL SETUP
This study set out to investigate the effectiveness of wireless diagnostics on the air conditioning compressor, by identifying the faulty.Hence a proper test experiment is needed.This research is done with vehicle and test rig which designed to replicate the actual vehicle automotive air conditioning system and powered by a power source supply replacing the usage of fuel combustion and also on the actual field testing vehicle.The compressor served as the primary function of the entire air conditioning system in the vehicle (Ahmad Z.A, et al. 2016, and Lee G.H. 2000).The air conditioning system is made up of several primary components, including a compressor, condenser, evaporator, thermal expansion valve, and receiver filter.Figures 1 and 2 depict the automobile air conditioning test rig and the typical vehicle A/C system, respectively.Matrix Laboratory or in its simple abbreviation MATLAB, is employed in the analysis of the data obtained to find the connection between the parameters set up during the experiment such as the percentage of oil in A/C, the volume of the refrigerant R134a, and last but not least the belting tension of the belt.Z-freq 2D is the coefficient generated from the original Z-freq coefficient, which is used to calculate the data from two axial points.Axis X is the data obtained from the horizontal direction of vibration while axis Y is the point of vibration that appears in the vertical direction, on the operating A/C compressor.Using a machine learning which a method used by a great deal of researcher in their research purposes, are able to process vast amount of data and can detects faults in machinery in this case faulty in the vehicle HVAC compressor whether the compressor operates in normal or abnormally (Mochammad S., et 4: Set of Parameters on Actual Car The effects of each of the parameters on the vibration signal were verified through the MATLAB analysis, and the parameters variables are shown in Figure 3, Figure 4 and Figure 5.Each of the experiments was conducted using experimental procedure of three types of variables, which as a constant variable, manipulated variable and a response variable as shown in Figure 3 for test rig and Figure 4 for actual vehicle HVAC.To ensure the experiment test are not interfere with unnecessary substance example old oil in the test rig, the rig is recovered, vacuumed and flushed by using a Robinair Refrigerant Machine.This procedure of recovery system is according to EPA standards and used by most automotive workshop as technicians use these devices before repairing or performing maintenance on vehicles.Before the experimentation begins, a recovery system is initiated as shown in Figure 9 for both test rig and vehicle.Recovery is essential to be conducted to recover all gases that present in the HVAC system using the Robinair machine which needs to be done each time servicing the HVAC system.Once the recovery process is complete, the vacuum process (Figure 10) is then initiated to ensure there is completely no trapped gas inside the system and then the flushing system is initiated once the vacuum process is completed.Flushing is very important to ensure there is no leftover oil left in the HVAC system and inside the compressor.This initial process usually took up to 1 hour of process.This is considered a calibration process for the HVAC system in the vehicle, which ensures the accuracy of data obtained is not influenced by any unnecessary substances in the system such as oil sediment in the compressor.The existence of leftover materials in the compressor or any parts of the vehicle HVAC system, will lead to noises and vibration in the vehicle, which need to be addresses as a significant issues, and this is a reason of need diagnose the system [ Zhiwei C, et al. 2018].

MATHEMATICAL MODELLING
The machine learning method is the mathematical and statistical technique that is very useful for the analysis of the malfunction or faulty in the A/C compressor which is related to any of the parameters, either lack of refrigerant or lack of oil in the system and et cetera.The method involving in using the coefficient formula to calculate and identify any data changes in the experiment analysis.Integrated Kurtosis Algorithm (I-Kaz), generally developed to detect any changes in the data signal in the experiment and measure the data collected concerning the data centroid of dynamic signal analysis.I-Kaz algorithm analysis provides a three-dimensional graphic display of the data collected with a different magnitude distribution of each axis, in which each axis represents the data distribution of frequency ranges [M.Z Nuawi, et al. 2008].Table 1 shows the normalized I-Kaz coefficient together with the Kurtosis coefficient and equation (1) shows the I-Kaz formula.
Where, N = number of data M4 = 4th order moment of signal in LF, HF and VF range respectively.In general, the formula used in statistical parameters is the mean value, the standard deviation value, the root mean square (RMS) value, the skewness and the kurtosis (Cain, M. K., et al. 2017).The statistical moment for kurtosis (K) is the fourth moment of global signal statistics which is sensitive to the spikiness of data.The kurtosis value is stated as follows in equation ( 2): Where, K: Kurtosis, n: The number of sample, f: frequency, σ: The Square Standard deviation value The kurtosis value is approximately 3.0 for a Gaussian distribution.Higher kurtosis values designated the occurrence of extreme values and established in a Gaussian distribution [M.Z Nuawi, et al. 2008].For a signal with an n-number of data points, the standard deviation value s is given by the following equation (3): Where, Xi: value of data point, μ: mean of the data, σ = 2 From the previous equation stated, the coefficient of Z-freq 2D is defined as the following equation (4).The Z frequency is calculated based on two points of data frequency, which are x represented by the x-axis location and y represented by the y-axis location.
Where,: 2 : z frequency 2D at second central moment, X: x-axis location, Y: y-axis location.

COMPARISON OF DATA RESULTS FOR VEHICLE AND TEST RIG
The figure 13 shows the data collection of the vibration signal, which is obtained by using the software Digivibes.The vibration signal that catches by the sensor device Phantom Vibration sensor, is sent to the cloud data gate which then was recorded and can be traced through the software as shown in figure 6.The symbol H is indicated for horizontal which means Z-axis, V is vertical for X-axis and A is axial for Y-axis.Each experimental test was conducted using a 3.2 kHz sampling rate and 6400 lines of resolution which produce about 16, 384 samples per channel in about 5s of duration of recording.The reason for using this method is so that every event that occurred during the capturing of the vibration signal can be detected and analyzed.For example, whenever the vehicle A/C hit a high temperature inside the evaporator the compressor will immediately work to reduce the temperature from about 9 degrees Celsius to 10 degrees Celsius.This causes the engine rpm to drop and need to be adjusted back to the respective rpm needed.During this event, the Phantom Vibration would detect all vibration data and the data can be isolated from drop rpm to actual rpm.
The data are collected according to the set of parameters in the experiment.Figure 14 presents the graph plot of amplitude vs time which is known as the time domain graph for the parameter of 320g volume of refrigerant R134a with 80ml A/C oil.The graph was then converted into frequency domain respectively to vertical and horizontal axial.From Table 2, we can see that as the engine speed increases from the idle speed of the vehicle which is 900 RPM to the max revolution per minute of 3000 RPM, the values of the Z-freq 2D are increasing.Indicated that as the RPM increase, the vibration that is present on the vehicle compressor are increasing as well.The engine speed of 900 RPM will give the result of 3.3355 of the Z-freq 2D coefficient.This is shown that the Phantom Vibration sensor is able to detect and determine the engine speed by monitoring the vibration and providing samples of graphs to be examined.Overall of the summary, 900 RPM will get 3.3355 Z-freq 2D, 1000 RPM is 7.9132, 1500 RPM set to be 12.6318, 2000 RPM is 87.434, 2500 RPM will get 200.0343 and last but not least 3000 RPM is 399.2113.The representation of data in Table 2 can be seen in Figure 21 for Z-freq 2D vs engine speed and Figure 22 for Z-freq 2D vs compressor speed.It can be concluded from the graph chart that the magnitude of the frequency increases significantly as the engine or the compressor speed increase which contributed to Z-freq 2D values.Vehicle engine or compressor data patterns can also be displayed in a scatter diagram as shown in Figure 15 until Figure 20.The scatter graph of the vibration shows the significant pattern of the vibration captured by the sensor.The x-axis represents of horizontal vibration frequency distribution while the y-axis represents the vertical vibration frequency distribution.It appears that the scatter pattern is becoming wider in both frequency axial as the speed of the engine or the compressor are increasing.In order to capture all the scatter patterns, the graph has been scale set up to 1000 m/s^2 for both axes.In the scatter pattern graph, the distribution is scatter wider aligned to both axial axes which is due to the frequency domain increase in frequency magnitude.In this experiment the temperature of the HVAC system is recorded from 900 RPM of the engine until 3000 RPM.There are four point of temperature measured by using the Thermometer device, at the inlet of the evaporator, between the evaporator and compressor, and the inlet and outlet of the condenser.The temperature is represent as the symbols T1, T2, T3, and T4. Figure 23 shows the position of the thermometer is placed.The data of the temperature of each point was shown in the device and recorded, starting from the idle speed of the engine 900 RPM, 1000 RPM, 1500 RPM, 2000 RPM, 2500 RPM and 3000 RPM.In most cases, the temperature of the tube is taken for the calculation of the Coefficient of Performance (COP).COP is used and serves as a method of measurement to evaluate and calculate the index to measure how well the air conditioning system in HVAC works.COP is also the ratio of the rate of net removal to the rate of total energy input, which indicated that the higher the COP values the less cost to operate, the better for the environment, less energy usage and it is more efficient (Sulaiman N., et al. 2020).However in this study, the temperature recorded are for monitoring purposes and not for COP calculation.The test rig experimental data is set to test at parameters of 300g of Refrigerant R134a with 150ml of A/C oil and the experiment is conducted at 3.2 kHz sampling rate in duration of 5s period of time for recording vibration signal.Data were obtained as shown in the respective figure below.Figure 29 represents the plot graph of amplitude vs time which is known as the time domain graph for the parameter of 300g volume of refrigerant R134a with 150ml A/C oil.The graph was then converted into frequency domain respectively to vertical and horizontal axial.From Table 4, we can see that as the compressor speed increases from 1000 RPM to the max revolution per minute of 1500 RPM, the values of the Z-freq 2D are increasing as well.Indicating that as the speed of the compressor increased, the Z-freq 2D coefficient increased as well along with the increasing of the vibration signal.Overall of the summary, 1000 RPM will get 631.3161Z-freq 2D, 1100 RPM is 1279.7699,1200 RPM set to be 967.1356,1300 RPM is 1919.7964,and last but not least 1400 RPM will get 7972.4995.The representation of data in Table 4 can be seen in Figure 35 for Z-freq 2D vs compressor speed.It can be concluded from the graph chart that the magnitude of the frequency increases significantly as the engine or the compressor speed increase which contributed to Z-freq 2D values.The scatter data pattern of Z-freq 2D for the test rig is shown in Figure 30 until Figure 34.From the scatter graph, the scatter pattern is becoming wider in both frequency axial as the speed of the compressor is increasing.The distribution of the scatter pattern is due to the frequency domain, the frequency data distribution slowly increase in its frequency magnitude.

Figure 1 :
Figure 1: Automotive air conditioning Test Rig

Figure 2 :
Figure 2: Vehicle Air Conditioning System

Figure 5 :
Figure 5: The manipulated values for each of variables

Figure 6 :
Figure 6: Robinair Refrigerant A/C Recovery The wireless sensor, Phantom Vibration Sensor is set up at the point of interest which is the A/C compressor as shown in Figure 7 and Figure 8.

Figure 9 :
Figure 9: Recovery system -High and Low hose on test rig

Figure 13 :
Figure 13: Data collection using software Digivibes MX The experiment of the vehicle begins with a set of parameters of 320g volume of refrigerant R134a with 80ml of A/C oil for the actual vehicle and compared with the data obtained for 300g volume of refrigerant R134a with 150ml amount of oil for the test rig and the test are conducted during compressor on.Data obtained respectively are shown in the figures below.

Figure 14 :
Figure 14: Time and frequency domain at 900 RPM with compressor on

Figure 23 :
Figure 23: Sketch diagram of Thermometer position in HVAC positioned inlet and outlet of both evaporator and condenser

Figure 29 :
Figure 29: Time and frequency domain at 1000 RPM with compressor on

Table 3 :
Temperature recorded data with the engine speed