4. Artificial neural network

There are two different basic ways of thinking: logic and intuition. Logical thinking refers to the process of reasoning according to logical rules; First, it transforms information into concepts and represents them with symbols. Then, logical reasoning is carried out in serial mode according to symbolic operation. This process can be written as a serial instruction for the computer to execute. Intuitive thinking is to synthesize distributed information, and the result is a sudden idea or a solution to the problem. The fundamental point of this way of thinking lies in the following two points: 1. Information is distributed on the network through the excitation mode on neurons; 2. Information processing is accomplished through the dynamic process of simultaneous interaction between neurons.

Artificial neural network is the second way to simulate human thinking. This is a nonlinear dynamic system, which is characterized by distributed storage and parallel collaborative processing of information. Although the structure of a single neuron is extremely simple and its function is limited, the behaviors that can be realized by a network system composed of a large number of neurons are extremely colorful.

4. 1 learning principle of artificial neural network

Artificial neural network must learn according to certain learning criteria before it can work. Taking the recognition of handwritten letters "A" and "B" by artificial neural network as an example, it is specified that "A" outputs "1" and "B" outputs "0".

Therefore, the rule of online learning should be: if the network makes a wrong judgment, the network should reduce the possibility of making the same mistake next time through online learning. Firstly, a random value is given to each connection weight in the interval of (0, 1), and the image pattern corresponding to "a" is input into the network. The network adds the input modes by weight, compares them with the threshold, and then performs nonlinear operation to get the output of the network. In this case, the probability that the network output is "1" and "0" is 50% respectively, which means it is completely random. At this time, if the output is "1" (the result is correct), the connection weight is increased, so that the network can still make a correct judgment when it encounters the "A" mode input again.

If the output is "0" (that is, the result is wrong), the network connection weight is adjusted in the direction of reducing the comprehensive input weight, so as to reduce the possibility that the network will make the same mistake next time it encounters the "A" mode input. With this operation adjustment, when several handwritten letters "A" and "B" are input into the network in turn, the correct rate of network judgment will be greatly improved after learning several times through the network according to the above learning method. This shows that the network has successfully learned these two modes and memorized them in every connection weight of the network. When the network encounters any of these modes again, it can make a quick and accurate judgment and identification. Generally speaking, the more neurons a network contains, the more patterns it can remember and recognize.

4.2 Advantages and disadvantages of artificial neural network

Artificial neural network simulates the organizational pattern of brain neurons and has some basic characteristics of human brain function, which opens up a new way for the research of artificial intelligence. Neural network has the following advantages:

(1) parallel distributed processing

Because the arrangement of neurons in artificial neural network is not chaotic, it is often layered or arranged in a regular order, and signals can reach the input ends of a group of neurons at the same time, which is very suitable for parallel calculation. At the same time, if each neuron is regarded as a small processing unit, the whole system can be a distributed computing system, which avoids the previous problems such as "matching conflict", "combination explosion" and "infinite recursion" and has fast reasoning speed.

(2) learnable habits

A relatively small artificial neural network can store a large amount of expert knowledge, and can simulate the real environment according to the learning algorithm, or use a sample guidance system (called teacher learning) or adaptive learning input (called teacher-free learning) to continuously learn automatically and improve the storage of knowledge.

(3) Robustness and fault tolerance

Because a large number of neurons and their interconnections have the ability of associative memory and associative mapping, the fault-tolerant ability of expert system can be enhanced, and the failure or error of a few neurons in artificial neural network will not seriously affect the overall function of the system. But also overcomes the problem of "narrow knowledge order" existing in traditional expert systems.

(4) Generalization ability

Artificial neural network is a large-scale nonlinear system, which provides the potential for self-organization and cooperation of the system. It can completely approximate the complex nonlinear relationship. When the input changes slightly, its output can keep a fairly small gap with the output produced by the original input.

(5) It has a unified internal knowledge representation, and any knowledge rules can be stored in the connection weights of the same neural network through learning examples, which is convenient for the organization and management of the knowledge base and has strong universality.

Although artificial neural network has many advantages, based on its inherent mechanism, it inevitably has its own weaknesses:

The most serious problem of (1) is that it cannot explain its own reasoning process and reasoning basis.

(2) Neural network can't ask necessary questions to users, and it can't work when there is insufficient data.

(3) Neural network turns the characteristics of all problems into numbers and all reasoning into numerical calculation, and the result is inevitably the loss of information.

(4) The theory and learning algorithm of neural network need to be further improved.

4.3 Development trend of neural network and its feasibility in diesel engine fault diagnosis

Neural network provides a brand-new theoretical method and technical realization means for state monitoring and fault diagnosis of modern complex large-scale systems. Neural network expert system is a new type of knowledge expression system, which is different from the high-level logic model of traditional expert system. It is a low-level numerical model, and information processing is carried out through a large number of simple interactions between processing elements (nodes). Because of its distributed information retention mode, it provides a new way for expert system to acquire, express and reason knowledge. It combines logical reasoning with numerical operation, and uses the learning function, associative memory function and distributed parallel information processing function of neural network to solve the problems of representation, acquisition and parallel reasoning of uncertain knowledge in diagnosis system. By learning the empirical samples, the expert knowledge is stored in the network in the form of weights and thresholds, and the imprecise diagnostic reasoning is completed by using the information retention of the network, which better simulates the reasoning process of experts based on experience and intuition rather than complex calculation.

However, this technology is an interdisciplinary field of knowledge and an immature subject. On the one hand, the failure of equipment is quite complicated; On the other hand, the artificial neural network itself still has many shortcomings:

(1) is limited by the existing research results of brain science. Due to the difficulty of physiological experiments, the understanding of the thinking and memory mechanism of the human brain is still superficial.

(2) A complete and mature theoretical system has not been established. At present, many artificial neural network models have been proposed. To sum up, these models are generally directed topological networks composed of nodes and their interconnections, and the matrix of interconnection strength between nodes can be established by some learning strategy. But this * * * alone is not enough to form a complete system. Most of these learning strategies are fragmented and cannot be unified in a complete framework.

(3) It has a strong strategic color. This is the natural result of solving some applications without the support of unified basic theory.

(4) The interface with traditional computing technology is immature. Artificial neural network technology can never completely replace traditional computing technology, and can only supplement it in some aspects, so it needs to further solve the interface problem with traditional computing technology in order to obtain its own development.

Although there are many shortcomings in artificial neural network at present, the intelligent fault diagnosis technology combining neural network with traditional expert system will still be the focus of future research and application. It gives full play to the advantages of both. Neural network is good at numerical calculation and suitable for shallow empirical reasoning; Expert system is characterized by symbolic reasoning, which is suitable for deep logical reasoning. The parallel operation of intelligent systems not only expands the scope of condition monitoring and fault diagnosis, but also meets the real-time requirements of condition monitoring and fault diagnosis. It emphasizes both symbolic reasoning and numerical calculation, so it can adapt to the basic characteristics and development trend of the current fault diagnosis system. With the continuous development and perfection of artificial neural network, it will be widely used in intelligent fault diagnosis.

According to the above advantages and disadvantages of neural network, it is a research trend to combine neural network with traditional expert system to establish a so-called neural network expert system. Theoretical analysis and application practice show that neural network expert system combines the advantages of both and has been widely studied and applied.

The structure and working principle of centrifugal refrigeration compressor are very similar to those of centrifugal blower. But its working principle is essentially different from that of piston compressor. It does not increase the steam pressure by reducing the cylinder volume, but depends on the change of kinetic energy. A centrifugal compressor has a working wheel with blades. When the working wheel rotates, the blades drive the steam to move or make the steam gain kinetic energy, and then part of the kinetic energy is converted into pressure energy, which increases the pressure of the steam. This kind of compressor is called centrifugal compressor because it continuously sucks refrigerant vapor and throws it out along the radial direction when working. Among them, according to the number of working wheels installed in the compressor, it can be divided into single-stage type and multi-stage type. If there is only one working wheel, it is called a single-stage centrifugal compressor, and if it is composed of several working wheels in series, it is called a multi-stage centrifugal compressor. In air conditioning, because of less pressurization, it is generally single stage, and most centrifugal refrigeration compressors used in other aspects are multi-stage. The structure of single-stage centrifugal refrigeration compressor is mainly composed of working wheel, diffuser and volute. When the compressor works, the refrigerant vapor enters the steam suction chamber axially from the steam suction port, and the refrigerant vapor from the evaporator (or intercooler) is guided uniformly into the high-speed rotating working wheel 3 (the working wheel is also called the impeller, which is an important part of the centrifugal refrigeration compressor, because only the working wheel can transfer energy to the steam) under the shunting action of the steam suction chamber. Under the action of the blades, the steam rotates at a high speed with the rotation of the work, and at the same time, due to the centrifugal force, it flows in the blade channel, so that the pressure and speed of the steam are improved. The steam coming out of the working wheel enters the diffuser 4 with a gradually expanding cross-sectional area (because the steam comes out of the working wheel at a high flow rate, the diffuser partially converts kinetic energy into pressure energy, thereby increasing the pressure of the steam). When steam flows through the diffuser, the speed decreases and the pressure increases further. The steam pas through that diffuser is collect in the volute and then guided to the intercooler or condenser through the exhaust port.

Second, the characteristics and characteristics of centrifugal refrigeration compressor

Compared with piston refrigeration compressor, centrifugal refrigeration compressor has the following advantages:

(1) single machine has large refrigeration capacity. When the refrigeration capacity is the same, it has small volume, small floor space and 5 ~ 8 times lighter weight than the piston.

(2) Because there are no wearing parts such as steam valve and piston ring, and there is no crank-connecting rod mechanism, it has the advantages of reliable work, stable operation, low noise, simple operation and low maintenance cost.

(3) There is no friction between the working wheel and the casing, so lubrication is not needed. Therefore, the refrigerant vapor does not contact with the lubricating oil, thereby improving the heat transfer performance of the evaporator and condenser.

(4) The adjustment of refrigeration capacity is economical and convenient, and the adjustment range is wide.

(5) The adaptability to refrigerants is poor, and the centrifugal refrigeration compressor of a certain structure can only adapt to one refrigerant.

(6) Because the refrigerant with relatively large molecular weight is applicable, it is only applicable to the refrigerant with large refrigeration capacity, generally above 250,000 ~ 300,000 kcal/hour ... If the refrigeration capacity is too small, it requires small flow, narrow flow passage, high flow resistance and low efficiency. However, after continuous improvement in recent years, the refrigeration capacity of a single centrifugal refrigeration compressor for air conditioning can be as small as 654.38+ million kcal/hour.

Relationship between refrigeration and condensation temperature and evaporation temperature.

According to physics, the change of the moment of momentum of a rotating body is equal to the external moment, then

T=m(C2UR2-C 1UR 1)

Multiply both sides by angular velocity ω, and you get

Tω=m(C2UωR2-C 1UωR 1)

That is to say, the external force n on the spindle is:

N=m(U2C2U-U 1C 1U)

Divide the two sides of the above formula by m to get the work done by the impeller on the unit mass of refrigerant vapor, that is, the theoretical energy head of the impeller. U2 C2

The characteristics of ω 2c2ur1R2ω1c1u1c2rβ centrifugal refrigeration compressor refer to the relationship between theoretical energy head and flow rate, and can also be expressed as refrigeration capacity.

w = U2C2U-u 1c 1U≈U2C2U

(Because the imported C 1U≈0)

C2U = U2-C2RCTGβC2R = Vυ 1/(A2υ2)

So there is

W= U22( 1-

Vυ 1

ctgβ)

2 2U2

Where: V is the volume flow of steam inhaled by the impeller (m3/s)

υ 65438+υ 2 —— steam specific volume at the inlet and outlet of the impeller (m3/kg) respectively.

A2, U2 —— outlet area (m2) and peripheral speed (m/s) of the outer edge of the impeller.

β-blade installation angle

As can be seen from the above formula, the theoretical energy head W is related to the compressor structure, rotating speed, condensation temperature, evaporation temperature and the volume flow of steam sucked by the impeller. U2, A2 and β are constants for a compressor with a certain structure and a certain speed, so the theoretical energy head W is only related to the flow rate V, evaporation temperature and condensation temperature.

According to the characteristics of centrifugal refrigeration compressor, refrigerant with relatively large molecular weight should be adopted. At present, the refrigerants used in centrifugal refrigerators are F- 1 1, F- 12, F-22, F-13, F- 1 10. At present, F- 1 1 and F- 12 are the most widely used centrifugal compressors for air conditioning in China. Usually, centrifugal refrigeration compressors are selected when the evaporation temperature is not too low and the refrigeration capacity is large. In addition, in the petrochemical industry, centrifugal refrigeration compressors use propylene and ethylene as refrigerants, and only centrifugal compressors with extremely large refrigeration capacity use ammonia as refrigerant.

Third, the adjustment of centrifugal refrigeration compressor

Centrifugal refrigeration compressor and other refrigeration equipment constitute a unified energy supply and consumption system. When the refrigeration unit is running, only when the refrigerant flow through the compressor is equal to the refrigerant flow through the equipment, and the energy drop generated by the compressor is adapted to the resistance of the refrigeration equipment, can the working condition of the refrigeration system remain stable. However, the load of the refrigerator always changes with the external conditions and the user's use of refrigeration capacity. Therefore, in order to meet the user's demand for cooling load change and safe and economic operation, it is necessary to adjust the refrigeration unit according to external changes. The adjustment of refrigeration capacity of centrifugal refrigeration unit includes: changing compressor speed1; 2. Adopt rotatable inlet guide vanes; 3. Change the water inlet of the condenser; 4 air intake throttle, etc., among which the most commonly used are rotary air intake guide vane adjustment and air intake throttle adjustment. The so-called rotating inlet guide vane adjustment is to rotate the guide vane at the inlet of the compressor, so that the steam entering the impeller produces a swirl, and the kinetic energy increased by the working wheel changes to adjust the refrigeration capacity. The so-called steam inlet throttling adjustment is to install a regulating valve on the steam inlet pipeline in front of the compressor. If you want to change the working condition of the compressor, adjust the size of the valve and reduce the pressure at the inlet of the compressor by throttling, thus adjusting the refrigeration capacity. The most economical and effective way to adjust the refrigeration capacity of centrifugal compressor is to change the angle of inlet guide vane, so as to change the speed direction (C 1U) and flow rate v of steam entering the impeller. However, the flow V must be controlled in a stable working range to avoid the efficiency decline.