METHOD FOR DETECTING ANOMALIES IN A CYCLIC MANUFACTURING PROCESS

Description

Technical field

[0001] The invention relates to a method for determining anomalies in a cyclic manufacturing process according to the preamble of the independent claim. State of the art

[0002] A cyclical manufacturing process is characterized by repetitive activities in the production of piece goods. The activities are completed in cycles and then repeated. By repeating the activities consistently, a large number of identical piece goods can be produced, which keeps manufacturing costs low. Examples of a cyclical manufacturing process are primary forming processes such as casting, injection molding, sintering, etc. The invention is described below using an injection molding process, which does not preclude the use of the invention in other primary forming processes.

[0003] In an injection molding process, piece goods are manufactured using an injection molding machine. Each injection molding machine has an injection mold with at least one cavity. A material is injected into the cavity under pressure. The injected material plasticizes and takes on the shape of the cavity. The thus formed material cools and forms a piece product. The cavity pressure in the cavity is an important parameter for the quality of the piece product. Anomalies in the cavity pressure, such as excessively high or too low cavity pressure, lead to rejects during the injection molding process, which increases manufacturing costs and must be avoided.

[0004] Document WO2017 / 144344A1 discloses a method for carrying out a cyclical manufacturing process in which the manufacturing process is adjusted using at least one process setting variable, and the manufactured piece goods must meet at least one quality characteristic to be considered good parts. Before the actual manufacturing process, the process setting variable is deliberately changed in an adjustment phase to determine whether good parts will continue to be produced with the changed process setting variable. For this purpose, process parameter variants are automatically measured. A process parameter zone is automatically formed from measured process parameter variants that lead to good parts. In the subsequent manufacturing process, it is automatically determined whether measured process parameters lie within the process parameter zone. As long as this is the case, good parts are produced, and quality control of the piece goods is no longer necessary.

[0005] The two objects of the present invention are to simplify and improve the method known from document WO2017 / 144344A1 for carrying out a cyclic manufacturing process. Description of the invention

[0006] At least one of the problems is solved by the features of the independent claim.

[0007] The invention relates to a method for determining anomalies in a cyclical manufacturing process; in which a piece of material is manufactured in one cycle of the manufacturing process; and in which cycle a mold cavity pressure is measured, wherein in the event of an anomaly in the mold cavity pressure, the piece of material is a defective part; wherein in a first step of the method, at least one temporal sequence of pressure values is provided for at least one cycle; wherein in a second step of the method, at least one characteristic value is automatically determined for each temporal sequence of pressure values; wherein in a third step of the method, for several cycles, the characteristic value is automatically partitioned in at least one decision tree and an average depth of the decision tree is automatically determined; wherein in a fourth step of the method, a temporal sequence of further pressure values is provided in a further cycle.wherein in a fifth step of the method, at least one further characteristic value is automatically determined for the temporal sequence of further pressure values; wherein in a sixth step of the method, the further characteristic value is automatically partitioned in at least one further decision tree and a further depth of the further decision tree is automatically determined; and wherein in a seventh step of the method, it is automatically determined whether, for a characteristic value and a further characteristic value with the same characteristic value index, the further depth of the further decision tree is less than the average depth of the decision tree and whether an anomaly in the cavity pressure exists.;

[0008] According to the invention, no process parameters are varied, as in document WO2017 / 144344A1, to create a stable process parameter zone where good parts are produced. Rather, for each cycle, the temporal sequence of pressure values of the cavity pressure is reduced to at least one parameter, which parameter is automatically partitioned for multiple cycles in a decision tree. This is because the inventive assumption is that a defective part originates from an anomaly in the cavity pressure and can be quickly partitioned in the decision tree with a shallow depth. Good parts, on the other hand, are much more similar by nature and therefore do not exhibit an anomaly in the cavity pressure. They cannot therefore be partitioned as quickly in the decision tree; their decision tree has a comparatively greater depth.According to the invention, only the further depth of a further decision tree for a further characteristic value from a current, further cycle needs to be compared with the average depth of a decision tree for the same characteristic value from several previous cycles in order to infer an anomaly in the cavity pressure in the current, further cycle from a depth difference. Thus, no quality control of the manufactured piece product is necessary in the current, further cycle to classify it as a good or bad part.

[0009] The invention is actually easier to implement than the method known from WO2017 / 144344A1. This is because the adjustment phase of deliberately changing a process setting variable is no longer necessary. Rather, in the third step of the method, the decision tree for partitioning the characteristic value of multiple cycles is automatically determined using unsupervised machine learning. Within the meaning of the invention, the adjective "automatic" means that a step of the method is carried out without human intervention. The third step of the method can thus be carried out simultaneously with the cyclical manufacturing process, which is advantageously simple because it saves time and effort.

[0010] Further developments of the subject matter of the invention are claimed in the dependent claims. Short description of the characters

[0011] In the following, the invention is explained in more detail using the figures as examples. Fig. 1 shows a flow chart with several steps V1 - V7 of the method for determining anomalies in a cyclic manufacturing process; Fig. 2 shows schematically components of an injection molding machine 1 for carrying out the method according to Fig. 1 ; Fig. 3 shows a molding machine 1 according to Fig. 2 determined cavity pressure curve Yk, k=1; Fig. 4 shows a cavity pressure curve Yk, k=1 determined with the injection molding machine 1 according to Fig.2 determined cavity pressure curve Yk, k=2 for a good part G and a cavity pressure curve Yk, k=3 for a bad part S; Fig. 5 shows a cavity pressure curve determined with the injection molding machine 1 according to Fig.2 determined cavity pressure curve Yk, k=2 for a good part G and a cavity pressure curve Yk, k=4 for a bad part S; Fig. 6 shows a decision tree Bi, i=1...n for carrying out the third step V3 of the method according to Fig. 1 ; and Fig. 7 shows a with the injection molding machine 1 according to Fig. 2determined further cavity pressure curve Y'; Fig. 8 shows a further decision tree Bi', i=1...n for carrying out the sixth step V6 of the method according to Fig. 1 and Fig. 9 shows a correlation table between at least one of the methods according to Fig. 1 determined characteristic value Kki, k=1...1 i=1...n and at least one machine setting variable Mh, h=1...r for the injection molding machine 1 according to Fig. 2 .

[0012] The same reference symbols refer to the same objects in the figures. Ways to implement the invention

[0013] Fig. 1 shows a flowchart with several steps V1 - V7 of the procedure for detecting anomalies in a cyclic manufacturing process.

[0014] The injection molding process is a cyclical manufacturing process with cyclically repeating activities with a plurality of cycles Zk, k=1...1, where the cycle index j denotes the individual cycles Zk, k=1...1 and the cycle number l denotes the number of cycles Zk, k=1...1. In each cycle Zk, k=1...1, a piece product 14 is produced and a cavity pressure P is measured.

[0015] In a first step V1 of the method, at least one temporal sequence of pressure values Xkj, k=1...1, j=1...m is provided for at least one cycle Zk, k=1...1.

[0016] In a second step V2 of the method, at least one characteristic value Kki, k=1...1, i=1...n is automatically determined for each temporal sequence of pressure values Xkj, k=1...l, j=1...m.

[0017] In a third step V3 of the method, the characteristic value Kki, k=1...1, i=1...n is automatically partitioned in a decision tree Bi, i=1...n and an average depth Ti, i=1...n of the decision tree Bi, i=1...n is automatically determined.

[0018] In a fourth step V4 of the method, a temporal sequence of further pressure values Xj', j=1...m is provided in a further cycle Z'.

[0019] In a fifth step V5 of the method, at least one further characteristic value Ki', i=1...n is automatically determined for the temporal sequence of further pressure values Xj', j=1...m.

[0020] In a sixth step V6 of the method, the further characteristic value Ki', i=1...n is automatically partitioned in a further decision tree Bi', i=1...n and a further depth Ti', i=1...n of the further decision tree Bi', i=1...n is automatically determined.

[0021] In a seventh step V7 of the method, it is automatically determined whether the further depth Ti', i=1...n of the further decision tree Bi', i=1...n is less than the average depth Ti, i=1...n of the decision tree Bi, i=1...n and whether there is an anomaly in the cavity pressure P.

[0022] The process can be carried out on any commercially available injection molding machine 1 known to the person skilled in the art. Fig. 2The schematically illustrated injection molding machine 1 has an injection device 11 in which a melt 12 is liquefied. The melt 12 can be made of plastic, metal, ceramic, etc. The injection molding machine 1 has an injection mold 13 with a cavity into which the liquefied melt 12 is injected under pressure by the injection device 11. The melt 12 injected into the cavity is caused to solidify in the cavity and cools. At the end of the manufacturing process, a finished piece 14 is ejected from the cavity.

[0023] At least one pressure sensor is arranged in the cavity, which measures a temporal progression of a cavity pressure P in the cavity. The pressure sensor is, for example, a piezoelectric pressure sensor, which generates a temporal sequence of pressure signals PSkj, k=1...1, j=1...m for the measured temporal progression of the cavity pressure P, where the signal index j denotes the individual pressure signals PSkj, k=1...1, j=1...m and the signal number m denotes the number of pressure signals PSkj, k=1...l, j=1...m. The pressure signals PSkj, k=1...1, j=1...m follow one another at times tj, j=1...m and are at a constant temporal interval from one another. The piezoelectric force transducer measures the cavity pressure P as electrical polarization charges. Each pressure signal PSkj, k=1...1, j=1...m is a set of electric polarization charges at a time tj, j=1...m.The amount of electrical polarization charges is proportional to the size of the cavity pressure P. The piezoelectric force transducer measures the cavity pressure P typically with a measurement accuracy of 1%, i.e. a cavity pressure P of 200 bar is measured with a measurement accuracy of + / - 2 bar. The piezoelectric force transducer measures the cavity pressure P with a temporal resolution of less than or equal to 0.01 Hz. This means that per unit time of 1 second the piezoelectric force transducer measures the cavity pressure P at least 100 times. For an injection molding process with a typical cycle time of 10 seconds the piezoelectric force transducer measures the cavity pressure P at least 1000 times and generates a temporal sequence of at least 1000 pressure signals PSkj, k=1...l, j=1...m.

[0024] The injection molding machine 1 has a quality control unit 15, which checks whether the manufactured piece product 14 meets at least one predefined quality characteristic. The predefined quality characteristic is a weight, dimensional accuracy, size, burr formation, cavity filling, burn mark, etc. If the manufactured piece product 14 meets the predefined quality characteristic, it is a good part G. If the produced piece product 14 does not meet the quality characteristic, it is a defective part S. The quality control unit 15 generates quality signals QS for the piece products 14 checked as good part G or defective part S.

[0025] The injection molding tool 13 has an evaluation unit 16, which has at least one converter unit 161, at least one computer 162, at least one input unit 163 and at least one output unit 164.

[0026] The evaluation unit 16 is connected to the injection molding tool 13 and the quality control unit 15 via signal lines. The evaluation unit 16 receives temporal sequences of pressure measurement signals PSkj, k=1...1, j=1...m from the injection molding tool 13 via the signal lines. Furthermore, the evaluation unit 16 receives quality signals QS from the quality control unit 15 via the signal lines.

[0027] To carry out the first step V1 of the method, the piezoelectric pressure sensor generates a temporal sequence of pressure signals PSkj, k=1...l, j=1...m in the form of electrical polarization charges. The converter unit 161 preferably has a charge amplifier. For each pressure signal PSkj, k=1...l, j=1...m, the charge amplifier converts the amount of electrical polarization charges into an electrical voltage value. The converter unit 161 then digitizes the electrical voltage value for each pressure signal PSkj, k=1...l, j=1...m into a value of the internal mold pressure P. Thus, the converter unit 161 automatically converts the temporal sequence of pressure signals PSkj, k=1...l, j=1...m into a temporal sequence of pressure values Xkj, k=1...l, j=1...m.

[0028] The computer 162 has at least one data processor and at least one data memory. At least one computer program C is stored in the data memory and can be loaded into the data processor. The computer program C loaded into the data processor causes the computer 162 to perform at least one step of the method. The computer 162 performs the step of the method automatically. For this purpose, the computer program C generates instructions.

[0029] The computer 162 can be operated via the input unit 163. The input unit 163 can be a keyboard for entering commands. In the context of the invention, the verb "operate" means that the computer 162 is started, controlled, and shut down by a person using commands via the input unit 163. The person is a human. The person is trained to operate the injection molding machine 1. The output unit 164 can be a screen on which digital data can be graphically displayed.

[0030] The computer program C loaded into the data processor instructs the computer 162 to load a time sequence of pressure values Xkj, k=1, j=1...m of the cycle Zk, k=1 into the data processor and to graphically display the loaded pressure values Xkj, k=1, j=1...m on the output unit 164 as a tool cavity pressure curve Yk, k=1, which is Fig. 3 The graphic representation has an ordinate and an abscissa. The ordinate denotes the cavity pressure P, and the ordinate denotes time t. The times tI to tIV and the cavity pressures PI to PIII and Pmax apply to the cavity pressure curve Yk, k=1. The injection mold 13 can have multiple cavities, and a specific cavity pressure curve Yk, k=1 can exist for each cavity.

[0031] Each cycle Zk, k=1...l has phases I to III, thus also the one in Fig. 3illustrated cycle Zk, k=1. An injection phase I begins at a time tI with an initial cavity pressure PI and ends at a time tII with a filling pressure PII. In the injection phase I, melt 12 is liquefied by a translational movement of a screw and moved against a nozzle. The liquefied melt 12 is injected through the nozzle into the cavity. For this purpose, the injection molding machine 1 has a control device 10 to control the liquefaction and injection of the melt 12 into the cavity. With the control device 10, at least one machine setting variable Mh, h=1...r such as a metering speed of the screw, an injection speed of the melt into the cavity, a switching time tII, a relief movement of the screw, a temperature of the melt, a temperature of the mold, etc. can be changed. The machine setting variable index h designates the individual machine setting variables Mh, h=1...r and the machine setting variable number r denotes the number of machine setting variables Mhi, h=1...r. For this purpose, the control device 10 has an input device for entering the machine setting variable Mh, h=1...r. The higher the injection speed, the faster the cavity is filled with melt. When the cavity is filled with melt 12, the cavity pressure P rises to a maximum cavity pressure Pmax in a short period of time. Shortly before the maximum cavity pressure Pmax is detected, the cavity is completely filled with melt 12, and the filling pressure PII is measured. The time tII at which the cavity is completely filled with melt is called the switching time tII. Injection phase I is ended.

[0032] At switchover time tII, the injection molding machine 1 is switched from the injection speed as the machine setting variable Mh, h=1...r to a holding pressure as the machine setting variable Mh, h=1...r; the holding pressure phase II begins. In the holding pressure phase II, the injection device 11 at the nozzle exerts a holding pressure on the melt 12 in the cavity. Shrinkage of the cooling melt 12 is also compensated for by additional melt 12 flowing into the cavity. In the process, the control device 10 reduces the cavity pressure P to a sealing point pressure PIII. The melt 12 is solidified in the cavity and cools down in the process. The holding pressure phase II begins at time tII and ends at a time tIII with the sealing point pressure PIII. The control device 10 can be used to change the holding pressure level and a holding pressure time as machine parameters M.The holding pressure level influences the cavity pressure P, particularly in the range of the maximum cavity pressure Pmax, where the melt 12 is briefly compressed and the cavity pressure P is greater than the filling pressure PII. The holding pressure time influences the rate at which the cavity pressure P decreases.

[0033] At time tIII, the residual cooling phase III begins; the solidified melt 12 continues to cool. Time tIII is also called the sealing point tIII, at which the melt 12 in the area of the nozzle has solidified to such an extent that no more melt 12 can flow into the cavity; the nozzle of the cavity is sealed. The injection molding tool 13 can be cooled using coolant. The control device 10 can be used to change a tool temperature as machine parameter M; the cavity is thus cooled to a greater or lesser extent in the holding pressure phase II and the residual cooling phase III. The residual cooling phase III ends at time tIV, at which the finished piece product 14 is ejected from the cavity.

[0034] The computer program C loaded into the data processor causes the computer 162 to perform the second step V2 of the method. To do so, the computer program C instructs the computer 162 to load the time sequence of pressure values Xkj, k=1, j=1...m into the data processor for the cycle Zk, k=1. The computer program C then instructs the computer 162 to automatically determine at least one characteristic value Kki, k=1, i=1...n for the time sequence of pressure values Xkj, k=1, j=1...m. The characteristic value index i designates the individual characteristic values Kki, k=1, i=1...n, and the characteristic value number n designates the number of characteristic values Kki, k=1, i=1...n.

[0035] A first characteristic value Kki, k=1, i=1 is the maximum cavity pressure Pmax of the cavity pressure curve Yk, k=1.

[0036] A second characteristic value Kki, k=1, i=2 is the time tmax of the maximum cavity pressure Pmax of the cavity pressure curve Yk, k=1.

[0037] A third characteristic value Kki, k=1, i=3 is the gradient of the cavity pressure P falling in the holding pressure phase II. The gradient is a partial derivative in a pressure value Xkj, k=1, j=1...m of the holding pressure phase II and therefore points in the direction of the largest change. In Fig. 3 the third characteristic value Kki, k=1, i=3 is shown on the cavity pressure curve Yk, k=1 as an arrow of the holding pressure phase II in the direction of the maximum cavity pressure Pmax.

[0038] A fourth characteristic value Kki, k=1, i=4 is the filling pressure PII on the cavity pressure curve Yk, k=1 in injection phase I.

[0039] A fifth characteristic value Kki, k=1, i=5 is the gradient of the cavity pressure P rising in the injection phase I until the switching time tII. The gradient is a partial derivative in a pressure value Xkj, k=1, j=1...m of the injection phase I and therefore points in the direction of the greatest change. In Fig. 3the fifth characteristic value Kki, k=1, i=5 is shown on the cavity pressure curve Yk, k=1 as an arrow of the injection phase I in the direction of the maximum cavity pressure Pmax.

[0040] A sixth characteristic value Kki, k=1, i=6 is the time tI at which the injection phase I begins.

[0041] A seventh characteristic value Kki, k=1, i=7 is the integral of the cavity pressure curve Yk, k=1 between the time tI at which the injection phase I begins and the time tmax of the maximum cavity pressure Pmax.

[0042] An eighth characteristic value Kki, k=1, i=8 is the gradient at the maximum cavity pressure Pmax. The gradient is a partial derivative and points in the direction of the largest change. Fig. 3 the eighth characteristic value Kki, k=1, i=8 is shown on the cavity pressure curve Yk, k=1 as a horizontal arrow in the maximum cavity pressure Pmax.

[0043] A new characteristic value Kki, k=1, i=9 is the gradient of the cavity pressure P falling in the residual cooling phase III. The gradient is a partial derivative in a pressure value Xkj, k=1, j=1...m of the residual cooling phase III and therefore points in the direction of the largest change. In Fig. 3 the ninth characteristic value Kki, k=1, i=9 is shown on the cavity pressure curve Yk, k=1 as an arrow of the residual cooling phase III in the direction of the maximum cavity pressure Pmax.

[0044] A ten-point characteristic value Kki, k=1, i=10 is the time tIV of the end of the residual cooling phase III.

[0045] An eleventh characteristic value Kki, k=1, i=11 is the time duration (tIV-tI) of the cycle Zk, k=1.

[0046] A twelfth characteristic value Kki, k=1, i=12 is an average cavity pressure Pm of the cycle Zk, k=1. The average cavity pressure Pm is the arithmetic mean of the pressure values Xkj, k=1, j=1...m of the cycle Zk, k=1.

[0047] A thirteenth characteristic value Kki, k=1, i=13 is an integral of the cavity pressure P over the time duration (tIV-tI) of the cycle Zk, k=1.

[0048] Fig. 4 shows a cavity pressure curve Yk, k=2 for a good part G and a cavity pressure curve Yk, k=3 for a bad part S. To distinguish between the two curves, the cavity pressure curve Yk, k=2 is shown with a dashed line, while the cavity pressure curve Yk, k=3 is shown with a solid line.

[0049] The cavity pressure curve Yk, k=2 originates from a temporal sequence of pressure signals PSkj, k=2, j=1...m generated by the piezoelectric pressure sensor in cycle Zk, k=2, which has been converted by the converter unit 161 into a temporal sequence of pressure values XGkj, k=2, j=1...m. The cavity pressure curve Yk, k=3 originates from a temporal sequence of pressure signals PSkj, k=3, j=1...m generated by the piezoelectric pressure sensor in cycle Zk, k=3, which has been converted by the converter unit 161 into a temporal sequence of pressure values XGkj, k=3, j=1...m.

[0050] The computer program C loaded into the data processor instructs the computer 162 to load the time sequences of pressure values Xkj, k=2, j=1...m and Xkj, k=3, j=1...m into the data processor for the cycle Zk, k=2, and the cycle Zk, k=3. The computer program C then instructs the computer 162 to automatically determine at least one characteristic value Kki, k=2, i=1...n for the time sequence of pressure values Xkj, k=2, j=1...m.

[0051] A fourteenth characteristic value Kki, k=2, i=14 is a sum of all pressure-normalized distances between pressure values Xkj, k=2, j=1...m of cycle Zk, k=2 with a good part G and pressure values Xkj, k=3, j=1...m of cycle Zk, k=3 with a bad part S. For the purposes of the invention, the adjective "pressure-normalized" means that the pressure value Xkj, k=2, j=1...m of cycle Zk, k=2 is taken as the reference. Each pressure-normalized distance is a straight line parallel to the ordinate between the pressure value Xkj, k=2, j=1...m of cycle Zk, k=2 taken as the reference and a pressure value Xkj, k=3, j=1...m of cycle Zk, k=3. The pressure-normalized distances between pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 and pressure values Xkj, k=3, j=1...m of the cycle Zk, k=3 are in Fig. 4 shown.

[0052] Fig. 5shows a cavity pressure curve Yk, k=2 for a good part G and a cavity pressure curve Yk, k=4 for a bad part S. To distinguish between the two curves, the cavity pressure curve Yk, k=2 is shown with a dashed line, while the cavity pressure curve Yk, k=4 is shown with a solid line.

[0053] The cavity pressure curve Yk, k=2 originates from a temporal sequence of pressure signals PSkj, k=2, j=1...m generated by the piezoelectric pressure sensor in cycle Zk, k=2, which has been converted by the converter unit 161 into a temporal sequence of pressure values XGkj, k=2, j=1...m. The cavity pressure curve Yk, k=4 originates from a temporal sequence of pressure signals PSkj, k=4, j=1...m generated by the piezoelectric pressure sensor in cycle Zk, k=4, which has been converted by the converter unit 161 into a temporal sequence of pressure values XGkj, k=4, j=1...m.

[0054] The computer program C loaded into the data processor instructs the computer 162 to load the time sequences of pressure values Xkj, k=2, j=1...m and Xkj, k=4, j=1...m into the data processor for the cycle Zk, k=2, and the cycle Zk, k=4. The computer program C then instructs the computer 162 to automatically determine at least one characteristic value Kki, k=2, i=1...n for the time sequence of pressure values Xkj, k=2, j=1...m.

[0055] A fifteenth characteristic value Kki, k=2, i=15 m is a sum of all time-normalized intervals between pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 with a good part S and pressure values Xkj, k=4, j=1...m of the cycle Zk, k=4 with a bad part S. In the sense of the invention, the adjective "time-normalized" means that two pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 and Xkj, k=4, j=1...m of the cycle Zk, k=4, which have the same time interval from the time tI at which the injection phase I begins, are related to each other. The time-normalized distance is a straight line between the pressure value Xkj, k=2, j=1...m of the cycle Zk, k=2 and the time-related pressure value Xkj, k=4, j=1...m of the cycle Zk, k=4. The time-normalized distances between pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 and pressure values Xkj, k=4, j=1...m of the cycle Zk, k=4 are in Fig. 5 shown.

[0056] With knowledge of the present invention, the person skilled in the art can use further parameters not described here.

[0057] Thus, for each cycle Zk, k=1...l, the temporal sequence of pressure values Xkj, k=1...l, j=1...m is reduced to at least one characteristic value Kki, k=1...1, i=1...n.

[0058] The computer program C instructs the computer 162 to store the cavity pressure curve Yk, k=1...1 in the data memory for each cycle Zk, k=1...1.

[0059] The computer program C instructs the computer 162 to store the determined characteristic value Kki, k=1...1, i=1...n in the data memory.

[0060] The computer program C loaded into the data processor causes the computer 162 to perform the third step V3 of the method. The computer program C instructs the computer 162 to automatically partition the at least one characteristic value Kki, k=1...1, i=1...n in at least one decision tree Bi, i=1...n for several cycles Zk, k=1...1 and to automatically determine an average depth Ti, i=1...n of the decision tree Bi, i=1...n. Within the meaning of the invention, the term "partitioning" means a complete separation of the characteristic value Kki, k=1...1, i=1...n of the several cycles Zk, k=1...1 into leaves B4i, i=1...n of the decision tree Bi, i=1...n.

[0061] Preferably, the partitioning of the characteristic value Kki (k=1...1, i=1...n) is performed by unsupervised machine learning using the decision tree Bi (i=1...n) for the graphical representation of hierarchically successive decisions. The adjective "unsupervised" indicates that the decision rules necessary for the decision are randomly selected by the computer 162 and are not formulated in advance. Fig. 6shows a decision tree Bi, i=1...n with a root node B1i, i=1...n, several paths B2i, i=1...n, several internal nodes B3i, i=1...n, and several leaves B4i, i=1...n. Preferably, the decision tree Bi, i=1...n is a binary tree, where from the root node B1i, i=1...n two paths B2i, i=1...n lead to two internal nodes B3i, i=1...n. The root node B1i, i=1...n is located on a first level of the decision tree Bi, i=1...n, the internal nodes B3i, i=1...n are located on a second level of the decision tree Bi, i=1...n. From each inner node B3i, i=1...n, two further paths B2i, i=1...n lead to at least one further inner node B3i, i=1...n or to at least one leaf B4i, i=1...n on further levels of the decision tree Bi, i=1...n.

[0062] The computer program C instructs the computer 162 to load at least one characteristic value Kkj, k=1...1, i=1...n into the data processor for several cycles Zk, k=1...1. The computer inputs the characteristic value Kki, k=1...1, i=1...n of the several cycles Zk, k=1...1 at a first level into the root node B1i, i=1...n. The computer program C instructs the computer 162 to apply at least one random decision rule E such as "greater than", "less than", "earlier than", "later than", etc., to the characteristic value Kki, k=1...1, i=1...n of the several cycles Zk, k=1...1 in the root node B1i, i=1...n. The decision rule E partitions the characteristic value Kki, k=1...1, i=1...n of the several cycles Zk, k=1...1. The application of decision rule E leads to a decision with two answers. According to Fig. 6the answers are two inner nodes B31, i=1...n below the root node B1i, i=1...n. Since the answers are further inner nodes B31, i=1, the computer program C instructs the computer 162 to apply another decision criterion E to the characteristic value Kki, k=1...1, i=1...n in the inner nodes B31, i=1...n and to continue the decision tree recursively. If the answer is a leaf B41, i=1...n, the characteristic value Kki, k=1...1, i=1 is partitioned.

[0063] The third step V3 of the method is explained below as an example. For a cycle number l=1000, the first characteristic value Kki, k=1...l000, i=1 is entered 1000 times into the decision tree Bi, i=1...n. This means that there are 1000 first characteristic values Kki, k=1...l000, i=1. Each first characteristic value Kki, k=1...l000, i=1 is a maximum cavity pressure Pmax, which was measured at a time tmax with a temporal resolution of 0.1 Hz and a measurement accuracy of 1%. For each cycle Zk, i=1...1000, there is a twin that consists of the maximum cavity pressure Pmax and the time tmax. The decision rule E is now applied to the 1000 twins, and the 1000 twins are partitioned into 1000 sheets B4i, i=1...n. The 1000 twins are completely separated. Each leaf B4i, i=1...n contains one twin with the first characteristic value Kki, k=1...1000, i=1.

[0064] The computer program C instructs the computer 162 to determine an average depth Ti, i=1...n of the decision tree Bi, i=1...n. To do so, the computer 162 determines an arithmetic mean of the number of levels of the decision tree Bi, i=1...n and outputs it as the average depth Ti, i=1...n of the decision tree Bi, i=1...n. The average depth Ti, i=1...n is Fig. 6 shown as a dotted line. Under the inventive assumption that an anomaly of the cavity pressure P can be partitioned quickly, an anomaly has a depth less than the average depth Ti, i=1...n. According to Fig. 6a characteristic value Kki, k=1...1, i=1...n is partitioned in a sheet B41, i=1 of the third level, the depth of this sheet B41, i=1 is less than the average depth Ti, i=1...n, from which it follows that the cycle Zk, k=1...1 from which the Kki, k=1...1, i=1...n in this sheet B41, i=1 of the third level comes, has an anomaly of the cavity pressure P.

[0065] Preferably, an entire forest with many decision trees Bi, i=1...n is determined. Through regression, a strengthened decision tree learns from previous decision trees. The previous decision trees are weak decision trees that could not isolate the characteristic value Kki, k=1...l, i=1...n at all, or they are strong decision trees that could not completely isolate the characteristic value Kki, k=1...1, i=1...n. Only a complete isolation, where each characteristic value Kki, k=1...l, i=1...n is partitioned into a leaf B4i, i=1...n, is a very good partitioning. The computer program C instructs the computer 162 to determine a strengthened decision tree Bi, i=1...n by selectively selecting and combining the strong decision trees Bi, i=1...n. The regressive determination of the reinforced decision tree Bi, i=1...n is continued in a large number of cycles l of cycles Zk, k=1...1 until a complete isolation of the characteristic value Kki, k=1...1, i=1...n.

[0066] Preferably, several characteristic values Kki, k=1...l, i=1...n are partitioned. This improves the quality of the partitioning. For the purposes of the invention, the term "quality" refers to the probability with which the partitioning of the characteristic value Kki, k=1...l, i=1...n actually distinguishes a good part G from a bad part S.

[0067] The partitioning of multiple characteristic values Kki, k=1...1, i=1...n is also explained, for example. For example, the separation of the first 1000 characteristic values Kki, k=1...1000, i=1 in the decision tree Bi, i=1...n may have been incomplete. Therefore, in a reinforced decision tree Bi, i=1...n, the first characteristic value Kki, k=1...1000, i=1 and the third characteristic value Kki, k=1...1000, i=3 are automatically partitioned 1000 times. Each first characteristic value Kki, k=1...1000, i=1 is a maximum cavity pressure Pmax measured at a time tmax. Every third characteristic value Kki, k=1...1000, i=3 is the gradient of the cavity pressure P decreasing in the holding pressure phase II. For each cycle Zk, i=1...1000, there is a triplet consisting of the maximum cavity pressure Pmax, the time tmax, and the gradient of the cavity pressure P decreasing in the holding pressure phase II. By applying decision rule E, the 1000 triplets are divided into 1000 sheets B4i, i=1...n partitioned. The 1000 triplets are completely separated; each leaf B4i, i=1...n contains a triplet with the first characteristic value Kki, k=1...1000, i=1 and with the third characteristic value Kki, k=1...1000, i=3.

[0068] The computer program C instructs the computer 162 to store the decision tree Bi, i=1...n with the characteristic value Kki, k=1...1, i=1...n partitioned for several cycles Zk, k=1...1 and the average depth Ti, i=1...n in the data memory.

[0069] In the fourth step V4 of the method, the piezoelectric pressure sensor generates a temporal sequence of further pressure signals PSj', j=1...m in a further cycle Z', which is converted by the converter unit 161 into a temporal sequence of further pressure values Xj', j=1...m.

[0070] The computer program C loaded into the data processor causes the computer 162 to perform the fifth step V5 of the method. To do so, the computer program C instructs the computer 162 to load the time sequence of further pressure values Xj', j=1...m, into the data processor for the further cycle Z'. The computer 162 graphically displays the loaded further pressure values Xj' j=1...m on the output unit 164 as a further cavity pressure curve Y'. The further cavity pressure curve Y' is in Fig. 7 shown.

[0071] Now the computer program C commands the computer 162 to automatically determine at least one further characteristic value Ki', i=1...n for the time sequence of further pressure values Xj', j=1...m.

[0072] A first additional characteristic value Ki', i=1 is the maximum cavity pressure Pmax of the further cavity pressure curve Y'.

[0073] A second additional characteristic value Ki', i=2 is the time tmax of the maximum cavity pressure Pmax of the further cavity pressure curve Y'.

[0074] A third additional characteristic value Ki', i=3 is the gradient of the cavity pressure P falling in the holding pressure phase II.

[0075] A fourth additional characteristic value Ki', i=4 is the filling pressure PII on the further cavity pressure curve Y' in the injection phase I.

[0076] A fifth further characteristic value Ki', i=5 is the gradient of the cavity pressure P increasing in the injection phase I up to the switching time tII.

[0077] A sixth further characteristic value Ki', i=6 is the time tI at which the injection phase I begins.

[0078] A seventh further characteristic value Ki', i=7 is the integral of the further cavity pressure curve Y' between the time tI with which the injection phase I begins and the time tmax of the maximum cavity pressure Pmax.

[0079] An eighth further characteristic value Ki', i=8 is the gradient at the maximum cavity pressure Pmax.

[0080] A further characteristic value Ki' i=9 is the gradient of the cavity pressure P falling in the residual cooling phase III.

[0081] A ten-fold further characteristic value Ki', i=10 is the time tIV of the end of the residual cooling phase III.

[0082] An eleventh further characteristic value Ki', i=11 is the duration (tIV-tI) of the further cycle Z'.

[0083] A twelfth additional characteristic value Ki', i=12, is an average cavity pressure Pm of the subsequent cycle Z'. The average cavity pressure Pm is the arithmetic mean of the additional pressure values Xj', j=1...m of the subsequent cycle Z'.

[0084] A thirteenth further characteristic value Ki', i=13 is an integral of the cavity pressure P over the time period (tIV-tI) of the further cycle Z'.

[0085] A fourteenth further characteristic value Ki', i=14 is a sum of all pressure-normalized distances between pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 and further pressure values Xj', j=1...m of the further cycle Z'.

[0086] A fifteenth further characteristic value Ki', i=15 is a sum of all time-normalized distances between pressure values Xkj, k=2, j=1...m of the cycle Zk, k=2 and further pressure values Xj', j=1...m of the further cycle Z'.

[0087] Thus, for the further cycle Z', the temporal sequence of further pressure values Xj', j=1...m is reduced to at least one further characteristic value Ki' i=1...n.

[0088] The computer program C commands the computer 162 to store the further cavity pressure curve Y' in the data memory.

[0089] The computer program C instructs the computer 162 to store the determined further characteristic value Ki', i=1...n in the data memory.

[0090] The computer program C loaded into the data processor causes the computer 162 to perform the sixth step V6 of the method. The computer program C instructs the computer 162 to automatically partition the at least one further characteristic value Ki', i=1...n, into at least one further decision tree Bi', i=1...n for the further cycle Z' and to automatically determine an average further depth Ti', i=1...n of the further decision tree Bi', i=1...n.

[0091] For this purpose, the computer program C commands the computer 162 to load at least one further characteristic value Kj', i=1...n, into the data processor for the next cycle Z'. According to Fig. 8the computer inputs the further characteristic value Ki', i=1...n at a first level into a further root node B1', i=1...n of the further decision tree Bi', i=1...n. The computer program C instructs the computer 162 to apply at least one random decision rule E to the further characteristic value Ki', i=1...n in the further root node B1', i=1...n. Using the decision rule E, the further characteristic value Ki', i=1...n is partitioned via further paths Bi2', i=1...n and further inner nodes Bi3', i=1...n in further leaves Bi4', i=1...n.

[0092] The computer program C further instructs the computer 162 to determine a further depth T', i=1...n of the further decision tree Bi', i=1...n. To do so, the computer 162 determines the number of levels of the further decision tree Bi', i=1...n. According to Fig. 8The further decision tree Bi', i=1...n has three levels. The computer 162 outputs the number of levels of the further decision tree Bi', i=1...n as the further depth Ti', i=1...n of the further decision tree Bi', i=1...n.

[0093] The computer program C instructs the computer 162 to store the further decision tree Bi', i=1...n with the further characteristic value Ki', i=1...n partitioned for the further cycle Z' and the further depth Ti', i=1...n in the data memory.

[0094] The computer program C loaded into the data processor causes the computer 162 to carry out the seventh step V7 of the method.

[0095] To this end, the computer program C instructs the computer 162 to load the average depth Ti, i=1...n of the decision tree Bi, i=1...n for the characteristic value Kki, k=1...l, i=1...n and the further depth T' of the further decision tree Bi', i=1...n for the further characteristic value Ki', i=1...n into the data processor. The characteristic value Kki, k=1...l, i=1...n and the further characteristic value Ki', i=1...n have the same characteristic index i. The computer 162 determines whether the further depth Ti', i=1...n of the further decision tree Bi', i=1...n is less than the average depth Ti, i=1...n of the decision tree Bi, i=1...n and whether an anomaly in the cavity pressure P exists. If there is an anomaly in the cavity pressure P, the piece good 14 is qualified as a defective part S; if there is no anomaly in the cavity pressure P, the piece good 14 is qualified as a good part G.

[0096] The decision tree Bi, i=1...n is not only suitable for detecting anomalies in the cavity pressure P, it is also suitable for classifying piece goods 14 as a good part or a bad part S. This is because, even with a slight anomaly in the cavity pressure P, the manufactured piece goods 14 can still be a good part G. Although the piece goods 14 is then qualified as a bad part S due to a small additional depth Ti', i=1...n, during an inspection in the quality control unit 15, the manufactured piece goods 14 still meets the predefined quality characteristic and is recognized as a good part G. Qualifying a bad part S due to a small additional depth Ti', i=1...n is therefore a "false positive"; in fact, the piece goods 14 is a good part G. Conversely, it may be that a piece goods 14 is classified as a bad part G due to a sufficiently large additional depth Ti', i=1...n is qualified as a good part G, the piece goods 14 then fail to meet the predefined quality characteristic during an inspection in the quality control unit 15 and are identified as a defective part S. The qualification as a good part G due to a sufficiently large additional depth Ti', i=1...n is therefore "false negative"; in fact, the piece goods 14 are a defective part G.

[0097] In order to increase the sensitivity of the decision tree Bi, i=1...n in classifying piece goods 14 as good part G or bad part S, the computer program C loaded into the data processor causes the computer 162 to optimize the average depth Ti, i=1...n of the decision tree Bi, i=1...n.

[0098] Upon detection of "false positive" qualified piece goods 14, the computer 162 determines an optimized depth T+i, i=1...n of the decision tree Bi, i=1...n, which consists of the average depth Ti, i=1...n and is shortened by a "false positive" factor α: T + i = Ti − α , i = 1 … n

[0099] Upon detection of "false negative" qualified piece goods 14, the computer 162 determines an optimized depth T+i, i=1...n of the decision tree Bi, i=1...n, which consists of the average depth Ti, i=1...n and is shortened by a "false negative" factor β: T + i = Ti − β , i = 1 … n

[0100] Preferably, the "false positive" factor α shortens the average depth Ti, i=1...n by 10%. Preferably, the "false negative" factor β lengthens the average depth Ti, i=1...n by 10%. The optimized depth T+i, i=1...n of the decision tree Bi, i=1...n is particularly advantageous in the production of expensive piece goods 14, as it keeps manufacturing costs as low as possible.

[0101] The computer program C instructs the computer 162 to store the optimized depth T+i, i=1...n of the decision tree Bi, i=1...n, the "false positive" factor α and the "false negative" factor β in the data memory.

[0102] The anomaly of the cavity pressure P is detected via a characteristic value Kki, k=1...1, i=1...n. Using expert knowledge, it is possible to assign at least one machine setting variable Mh, h=1...r for injection molding machine 1 to the at least one characteristic value Kki, k=1...1, i=1...n in order to correct the anomaly of the cavity pressure P. The machine setting variable Mh, h=1...r is stored in the data memory.

[0103] A first machine setting variable Mh, h=1 is the switching time tII. The switching time tII can be adjusted via a defined position of the screw in its translational motion.

[0104] A second machine setting variable Mh, h=2 is a pressure in a hydraulic system of the screw in the holding pressure phase II. This can be adjusted.

[0105] A third machine setting variable, Mh, h=3, is the metering speed of the screw, i.e., the speed at which the screw moves the melt 12 toward the nozzle. This can be adjusted.

[0106] A fourth machine setting, Mh, h=4, is the injection speed of the melt through the nozzle into the cavity. This can be adjusted.

[0107] A fifth machine setting variable Mh, h=5 is the relief movement of the screw in injection phase I. This can be adjusted.

[0108] A sixth machine setting, Mh, h=6, is the melt temperature. This can be adjusted.

[0109] A seventh machine setting, Mh, h=7, is the tool temperature. This can be adjusted.

[0110] Fig. 9 shows a correlation table between the method according to Fig. 1determined at least one characteristic value Kki, k=1...1, i=1...n and at least one machine setting variable Mh, h=1...r for the injection molding machine 1 according to Fig. 2 The strength of the correlation between the characteristic value and the machine setting variable Mh, h=1...r varies on a scale from -1.0 (weak) to 0.0 (neutral) to +1.0 (strong). A machine setting variable Mh, h=1...r can be identified for a characteristic value Kki, k=1...1, i=1...n based on the strength of the correlation.

[0111] Upon detection of an anomaly in the cavity pressure P, the computer program C loaded into the data processor causes the computer 162 to identify at least one machine setting variable Mh, h=1...r, which exhibits a pronounced correlation with the at least one characteristic value Kki, k=1...1, i=1...n determined in the second step V2 of the method, via the correlation table and to load it from the data memory. The at least one machine setting variable Mh, h=1...r with the strongest correlation is displayed on the output unit 164. For this purpose, a setting to be made for the machine setting variable Mh, h=1...r is also displayed on the output unit 164. The person thus receives a notification that an anomaly in the cavity pressure P exists and which setting of a machine setting variable Mh, h=1...r is to be made to correct the anomaly in the cavity pressure P. List of reference symbols

[0112] 1Injection molding machine 10Control unit 11Injection device 12Melt 13Injection mold 14Piece goods 15Quality control unit 16Evaluation unit 161Converter unit 162Computer 163Input unit 164Output unit α"false positive" factor β"false negative" factor BiDecision tree B1iRoot B2iPath B3iInner node B4iLeaf Bi'Further decision tree CComputer program EDecision criterion GGood part hMachine setting index iCharacteristic value index IInjection phase IIHolding pressure phase IIIResidual cooling phase jSignal index kCycle index KkiCharacteristic value lCycle number MhMachine setting mSignal number nCharacteristic value number PCavity pressure PIInitial cavity pressure PIIFilling pressure PIIISealing point pressure PSkjPressure signal PSj'additional pressure signal Pmaverage cavity pressure Pmaxmaximum cavity pressure QSquality signal rmachine setting variable number Sbad part Shisetting variable Tiaverage depth Ti'additional depth T+ioptimized depth ttime tIBeginning of the injection phasetIISwitchover time tIIISealing point tivEnd of the residual cooling phase tmaxTime of maximum cavity pressure XkjPressure value YkCavity pressure curve Y'Further cavity pressure curve ZkCycle

Claims

1. A method for detecting anomalies in a cyclic manufacturing process; wherein an item (14) is produced in one cycle (Zk, k=1...1) of the manufacturing process; and in which cycle (Zk, k=1...1) a cavity pressure (P) is measured; wherein, if there is an anomaly in cavity pressure (P), the item (14) is a bad part (S); wherein in a first step (V1) of the method at least one chronological sequence of pressure values (Xkj, k=1...1, j=1...m) for at least one cycle (Zk, k=1...1) is provided; wherein in a second step (V2) of the method for each chronological sequence of pressure values (Xkj, k=1...1, j=1...m) at least one characteristic value (Kki, k=1...1, i=1...n) is automatically determined; wherein in a third step (V3) of the method the characteristic value (Kki, k=1...1, i=1...n) for a plurality of cycles (Zk, k=1...1) is automatically partitioned in at least one decision tree (Bi, i=1...n) and an average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) is automatically determined; wherein in a fourth step (V4) of the method a chronological sequence of further pressure values (Xj', j=1...m) in a further cycle (Z') is provided; wherein in a fifth step (V5) of the method at least one further characteristic value (Ki', i=1...n) is automatically determined for the chronological sequence of further pressure values (Xj', j=1...m); wherein in a sixth step (V6) of the method the further characteristic value (Ki', i=1...n) is automatically partitioned in at least one further decision tree (Bi', i=1...n) and a further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is automatically determined; and wherein in a seventh step (V7) of the method it is automatically determined whether for a characteristic value (Kki, k=1...1, i=1...n) and a further characteristic value (Ki', i=1...n) having the same characteristic value index (i) the further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is lower than the average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) and whether an anomaly in cavity pressure (P) exists.

2. The method according to claim 1, characterized in that in the second step (V2) of the method for each chronological sequence of pressure values (Xkj, k=1...1, j=1...m) a plurality of characteristic values (Kki, k=1...1, i=1...n) is automatically determined; and that in the third step (V3) of the method for a plurality of cycles (Zk, k=1...1) a plurality of characteristic values (Kki, k=1...1, i=1...n) are automatically partitioned in at least one decision tree (Bi, i=1...n) and an average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) is automatically determined.

3. The method according to claim 2, characterized in that in the fifth step (V5) of the method a plurality of further characteristic values (Ki', i=1...n) are automatically determined for the chronological sequence of further pressure values (Xj', j=1...m); and in that in the sixth step (V6) of the method the plurality of further characteristic values (Ki', i=1...n) are automatically partitioned in at least one further decision tree (Bi', i=1...n) and a further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is automatically determined.

4. The method according to claim 3, characterized in that in the seventh step (V7) of the method it is automatically determined whether for the plurality of characteristic values (Kki, k=1...1, i=1...n) and the plurality of further characteristic values (Ki', i=1...n) having the same characteristic value index (i) the further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is lower than the average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) and whether there is an anomaly in cavity pressure (P).

5. The method according to any of the claims 1 to 4, characterized in that in the second step (V2) of the method at least one of the following characteristic values (Kki, k=1...1, i=1...n) is automatically determined: a maximum cavity pressure (Pmax) of a cavity pressure curve (Yk, k=1...1); a time point (tmax) of the maximum cavity pressure (Pmax) of a cavity pressure curve (Yk, k=1...1); a gradient of the cavity pressure (P) decreasing in a holding pressure phase (II); a filling pressure (PII) on a cavity pressure curve (Yk, k=1...1) in an injection phase (I); a gradient of the cavity pressure (P) increasing in an injection phase (I) up to a switchover time point (tII); a time point (tI) at which an injection phase (I) starts; an integral of a cavity pressure curve (Yk, k=1...1) between a time point (tI) at which an injection phase (I) starts and a time point (tmax) of a maximum cavity pressure (Pmax); a gradient at a maximum cavity pressure (Pmax); a gradient of the cavity pressure (P) decreasing in a residual cooling phase (III); a time point (tIV) of the end of a residual cooling phase (III); a time period (tIV-tI) of the cycle (Zk, k=1...1); a mean cavity pressure (Pm) of the cycle (Zk, k=1...1); an integral of the cavity pressure (P) over a time period (tIV-tI) of the cycle (Zk, k=1...1); a sum of pressure-normalized intervals between pressure values (Xkj, k=1...1, j=1...m) of a cycle (Zk, k=1...1) that produces a good part (S) and pressure values (Xkj , k=1...1, j=1...m) of a cycle (Zk, k=1...1) that produces a bad part (S); and a sum of time-normalized intervals between pressure values (Xkj, k=1...1, j=1...m) of a cycle (Zk, k=1...1) that produces a good part (S) and pressure values ( Xkj, k=1...1, j=1...m) of a cycle (Zk, k=1...1) that produces a bad part (S) .

6. The method according to claim 5, characterized in that in the fifth step (V5) of the method at least one of the following further characteristic values (Ki', i=1...n) is automatically determined: a maximum cavity pressure (Pmax) of a further cavity pressure curve (Yk'); a time point (tmax) of the maximum cavity pressure (Pmax) of a further cavity pressure curve (Y'); a gradient of the cavity pressure (P) decreasing in a holding pressure phase (II); a filling pressure (PII) on a further cavity pressure curve (Y') in an injection phase (I); a gradient of the cavity pressure (P) increasing in an injection phase (I) up to a switchover time point (tII); a time point (tI) at which an injection phase (I) starts; an integral of a further cavity pressure curve (Y') between a time point (tI) at which an injection phase (I) starts and a time point (tmax) of a maximum cavity pressure (Pmax); a gradient at a maximum cavity pressure (Pmax); a gradient of the cavity pressure (P) decreasing in a residual cooling phase (III); a time point (tIV) of the end of a residual cooling phase (III); a time period (tIV-tI) of the further cycle (Z'); a mean cavity pressure (Pm) of the further cycle (Z'); an integral of the cavity pressure (P) over a period of time (tIV-tI) of the further cycle (Z'); a sum of pressure-normalized intervals between pressure values (Xkj, k=1...1, j=1...m) of the cycle (Zk, k=1...1) and further pressure values (Xj', j=1...m) of the further cycle (Z'); and a sum of time-normalized intervals between pressure values (Xkj, k=1...1, j=1...m) of the cycle (Zk, k=1...1) and further pressure values (Xj', j=1...m) of the further cycle (Z').

7. The method according to claim 1, characterized in that the item (14) is classified as a bad part (S) if an anomaly in the cavity pressure (P) exists; and in that the item (14) is classified as a good part (G) if no anomaly in cavity pressure (P) exists.

8. The method according to claim 7, characterized in that when items (14) classified as "false positive" are detected, the average depth (Ti, i=1...n) is reduced by a "false positive" factor (α); and / or in that when items (14) classified as "false negative" are detected, the average depth (Ti, i=1...n) is reduced by a "false negative" factor (β).

9. The method according to any of the claims 1 to 8, characterized in that when an anomaly in the cavity pressure (P) is detected at least one machine setting variable (Mh, h=1...r) is identified that is strongly correlated with the at least one characteristic value (Kki, k=1...1, i=1...n) determined in the second step (V2) of the method; and in that an adjustment to be made of the machine setting variable (Mh, h=1...r) is output in order to eliminate the anomaly in the cavity pressure (P).

10. An injection molding machine (1) for carrying out the method according to any of the claims 1 to 9, in which anomalies in a cyclic manufacturing process are detected; wherein an item (14) is produced in one cycle (Zk, k=1...1) of the manufacturing process; and in which cycle (Zk, k=1...1) a cavity pressure (P) is measured; wherein, if there is an anomaly in cavity pressure (P), the item (14) is a bad part (S); wherein in a first step (V1) of the method at least one chronological sequence of pressure values (Xkj, k=1...1, j=1... m) for at least one cycle (Zk, k=1...1) is provided; wherein in a second step (V2) of the method for each chronological sequence of pressure values (Xkj, k=1...1, j=1...m) at least one characteristic value (Kki, k=1...1, i=1...n) is automatically determined; wherein in a third step (V3) of the method the characteristic value (Kki, k=1...1, i=1...n) for a plurality of cycles (Zk, k=1...1) is automatically partitioned in at least one decision tree (Bi, i=1...n) and an average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) is automatically determined; wherein in a fourth step (V4) of the method a chronological sequence of further pressure values (Xj', j=1...m) in a further cycle (Z') is provided; wherein in a fifth step (V5) of the method at least one further characteristic value (Ki', i=1...n) is automatically determined for the chronological sequence of further pressure values (Xj', j=1...m); wherein in a sixth step (V6) of the method the further characteristic value (Ki', i=1...n) is automatically partitioned in at least one further decision tree (Bi', i=1...n) and a further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is automatically determined; and wherein in a seventh step (V7) of the method it is automatically determined whether for a characteristic value (Kki, k=1...1, i=1...n) and a further characteristic value (Ki', i=1...n) having the same characteristic value index (i) the further depth (Ti', i=1...n) of the further decision tree (Bi', i=1...n) is lower than the average depth (Ti, i=1...n) of the decision tree (Bi, i=1...n) and whether an anomaly in cavity pressure (P) exists; characterized in that the injection molding machine (1) comprises a pressure sensor, which pressure sensor generates a chronological sequence of pressure signals (PSkj, k=1...1, j=1...m) for carrying out the first step (V1) of the method; in that the injection molding machine (1) comprises a converter unit (161), which converter unit (161) automatically converts the chronological sequence of pressure signals (PSkj, k=1...1, j=1...m) into a chronological sequence of pressure values (Xkj, k=1...1, j=1...m).

11. The injection molding machine (1) according to claim 10, characterized in that the injection molding machine (1) comprises at least one computer (162); in that said computer (162) comprises at least a data processor and at least a data memory; in that at least one computer program (C) is stored in the data memory and can be loaded into the data processor; in that the computer program (C) loaded into the data processor causes the computer (162) to carry out the second step (V2) of the method, the third step V3) of the method, the fifth step (V5) of the method, the sixth step (V6) of the method, and the seventh step (V7) of the method.