Triggering and Characterisation of Realistic Internal Short Circuits in Lithium-Ion Pouch Cells – A New Approach Using Precise Needle Penetration

: The internal short circuit (ISC) in lithium-ion batteries is a serious problem because it is


Introduction
The lithium-ion battery (LIB) is the favoured energy storage device for a variety of applications today [1,2].However, the safe operation of LIBs is still a major challenge [3,4].Due to the limited thermal stability of the cell materials used in LIBs, an insufficient heat dissipation can lead to a so-called Thermal Runaway (TR) of the cell, which means an uncontrolled, unstoppable temperature rise and the release of large amounts of toxic and explosive gases [5][6][7][8].Anyhow, the uniform definition of TR in LIBs is challenging due to the complex and individual characteristics and is the subject of further research [9][10][11].Already a small part (<12 %) of the released energy is needed to heat up a neighbouring cell to such an extent that it is also run into a TR [12].This Thermal Propagation (TP) can lead to the complete burnout of a battery system and is particularly feared because of the huge hazard for people and the environment [6,13].A wide variety of electrical (e.g.Table 1.The special standing of the ISC in the field of LIB safety.Aspects that are particularly addressed in this paper are highlighted by bold font.

High relevance for application
Unknown ISC characteristics Limited options for fault containment • Responsible for 38 % of electric vehicle fires between 2014 and 2019 [19] • ISCs caused by manufacturing defects responsible for 90 % of all field failures [20] • Most frequently occurring failure case according to Lai et al. [21] • Most important failure case of LIBs according to Liu et al. [22] • Many field failure events remain unexplainable [15,21,23] • Detection and root-cause analysis hindered by characteristic selfdestruction [24,25] • Shortcomings of available trigger methods for field-failure replication [10,26] • No -or only to a limited extent -replication of realistic ISC in standards for product certification [27,28] • Spontaneous appearance within approved operating limits [29] • Local containment of failure without visual indication from outside [29,30] • Difficult and complex timely fault detection [31,32] • Limited effectiveness of protection concepts [33] overcharging), thermal (e.g.external heat) or mechanical (e.g.crash) failures are possible causes of TR [14][15][16][17][18].
Another important failure type is the internal short circuit (ISC), which is characterised by the development of a cell-internal electroconductive contact between both electrodes.The ISC has been in special focus of battery safety for many years due to the high percentage of caused field-failures, the lack of knowledge about the failure causes as well as the failure characteristics and the great challenge for risk reduction strategies, such as early failure detection.Please refer to Table 1 for a brief literature review of this special position of the ISC.
ISC can result from a wide variety of sources [6,23,24,34] as listed in the following.Please refer to the given references for more detailed information.

•
Mechanical penetration of the separator due to penetration by sharp external objects [34,35].
However, the cause of many field failures remains unknown [15,21].The reason for this is that for various interest groups (e.g.users, manufacturers, insurances, etc.) mainly the particularly safety-critical failure scenarios (TR and TP) catch attention.Research into the causes of burnt-out cells (TR) is challenging due to the high irreversible destruction level and results often in only a minimum of new insights [24].In addition, abnormal cells that are largely intact and interesting for analysis are rarely available for detailed investigation [26,54].Consequently, it is of great interest to replicate application-relevant failure scenarios and to analyse the resulting damage characteristics in detail [15].For a comprehensive insight, cells in which no TR has occurred must also be examined.Only with a deep understanding of the failure progression is it possible to develop methods for intelligent failure detection [55].The controlled creation of realistic fault data becomes especially useful for the validation of such methods since there is no homogenous test case, yet [56].
The individual ISC behaviour results from numerous influencing factors.Each electrode consists of a collector and the matching active material.As a result, four different contact conditions for an ISC are possible [57].Due to the different material properties, such as the electrical and thermal conductivity, the melting temperature of the collector and the exothermic reaction behaviour of the active materials, each contact condition has an individual criticality [21,[57][58][59].Here, the contact between the Al-Collector of the cathode with the graphite active material (An) of the anode (Al↔An) is considered as the most critical failure case [30,34,58,60].
A contact between the two active materials of cathode (Ca) and anode (Ca↔An) is rated as very divergent concerning the criticality by the literature and must be further investigated [34,57].The resulting contact resistance is responsible for the criticality of the ISC and is influenced in particular by the size of the contact area [61,62], the material composition [62], the contact pressure [22,57] and the available amount of electrolyte [63].This indicates that the contact resistance is subject to a certain degree of scattering and thus represents the basis for an individual ISC behaviour [57,63].In cells, the typical small contact areas lead to a characteristic local fusing process with the first contact (corresponds to the 1 st ISC), which causes to a dynamic change of the ISC resistance [62].This produces an individual ISC behaviour [23], which can be classified according to the voltage behaviour to a specific ISC type, as shown in Figure 1.
In the case of the minor ISC (type I), the current flow is interrupted early so that the local temperature rise (∆T < 10 • C) and the voltage drop (∆U < 0.05 V according to [34]) is low.Due to the small amount of removed charge, the voltage relaxes to the initial state within a short time (t < 0.5 s), but a structurally weakened area remains, from which renewed ISCs can possibly arise [62].If, on the other hand, greater local heat input occurs, a stronger major ISC (type II) develops [64] with a more severe voltage drop (e.g.∆U > 0.1 V according to [34]) and a significant temperature increase (∆T > 10 • C).The further failure development of this ISC type II can be further divided into three subtypes (IIa, b and c), as shown in Figure 1.In type IIa, the dynamic short-circuit development is followed by the interruption of the ISC current, which leads to a relaxation of the cell voltage and decreasing of local temperatures.Due to the damage within the cell, the self-discharge is low but quite constant.In contrast, in type IIb a continuous dynamic ISC develops, which is characterised by periods with stronger and weaker voltage drops.Finally, the TR of the cell occurs, and the voltage drops abruptly to 0 V.A more violent cell reaction and an accelerated voltage drop to 0 V is characterised as type IIc.In particular, the combination of exothermic side reactions [64], the local evaporation of electrolyte [65] and the melting of the separator [22,23] lead to an individual and hard-to-predict ISC behaviour of type II.Furthermore, it is known that the state of charge (SOC) has a significant influence on the failure progression [22,66,67].
The replication of a massive ISC type IIc is comparatively simple, and can be realised by a multi-layer nail penetration or a high external loading [68][69][70].However, due to the intense rapid reaction, the nail test is more suitable for investigating the general reactivity of a cell or as a failure case for studying TP behaviour than to replicate field-relevant local ISCs.In many failure cases in practice, in particular ISC caused by dendrites or impurity particles, however, a very local single-layer ISC results [26,41].Although it has rarely been examined until now, especially on larger cells, a slower, initially less critical failure progression can be assumed here.As a result, these ISC types IIa and IIb, as well as the minor ISC type I, enable an early detection of failures and the initiation of suitable countermeasures to minimise the potential risk.But the challenge is to reproduce and characterise these local failures [26].Approaches exist to specifically prepare the cells and insert small components which subsequently initiate an ISC through an external temperature impact [71][72][73].Besides the increased preparation effort, it has been shown that these components are not always reliably activated and thus do not result in reproducible mechanical initial conditions for ISC development [43,74].In contrast, external penetration with a thin needle generates an ISC reliably [63,75].
This study has the goal to further optimise the needle penetration approach for replicating application-relevant ISCs and to demonstrate the possibilities of reproducible ISC initiation.An ISC leads to internal structural damage, which, in the form of type I or IIa, makes the cell appear quite inconspicuous for a long time [22].The failure detection of such an ISC is challenging, especially in the case of large-format cells [33,76], as a result of which it must be assumed that in practice not all failure cases are detected when the first failure occurs and consequently such damaged cells remain in operation [29].Various strategies can be considered for the detection of failures, whereby the change of the electrochemical behaviour in the form of voltage deviations is a particularly promising approach [77][78][79].This is another reason to analyse the electrochemical cell behaviour of cells with former or still existing ISC extensively in this work.Hereby, an essential contribution is made for a better understanding of the ISC and forms the basis for an enhanced failure detection and risk assessment.

Structure
To replicate close-to-reality ISCs, a new method was developed.The new approach consists of precise needle penetration, which leads to controlled mechanical ISC conditions.In Section 1.2, the initial thoughts on the development of the method are described.The used material is then introduced in Section 2. Here, the cell type (Sec.2.1), including the surrounding cell conditions (Sec.2.2), is described comprehensively.Furthermore, the temperature measurement (Sec.2.3) and the used penetration needle (Sec.2.4) are specified.The experimental set-up, including the measurement technique for the ISC characterisation, is then examined in the next Section 2.5.
Subsequently, the specific experimental procedure of ISC initiation is explained in Section 3.1.For a more detailed characterisation of the dynamic ISC behaviour, the essential key features of the ISC are introduced in Section 3.2.In Section 3.3, the procedure for post-characterisation of the damaged cells is described.
The results are discussed in Section 4. This is divided into the general ISC characteristics (Sec.4.1) and the investigation of the influence of penetration depth on ISC behaviour (Sec.4.2).Following this, the results of the short-term characterisation are summarised in Section 4.3.The results of the post-characterisation are separated in the electrochemical post-characterisation (Sec.4.4) and the post-abuse analysis (Sec.4.5).Finally, the main conclusions are outlined in Section 5.

General initial considerations
To replicate local, single-layer ISC by penetrating with a needle, it is necessary to insert a thin, sharp needle precisely and slowly into the cell.In previous studies, after the 1 st ISC, the needle penetration gets continuously deeper, which changes the primary initiation state [65,75].Concerning ISCs triggered by dendrites or impurities, the state where no further penetration occurs and the progression of ISCs remains self-sustaining is of particular interest.Due to the immediate stop of the needle penetration, the smallest possible contact area is created.The high requirements for the accuracy of the penetration are realised by a self-developed test rig (Sec.2.5).To investigate a better contact condition, it is useful to realise deeper penetrations as a comparison (Sec.4.2).However, ISC behaviour can only be correctly rated if the exact mechanical initiation condition is known.This requires experiments with a dummy cell.(Sec.4.1.1).The low reproducibility of triggers for replicating an ISC is generally criticised, and thus complicates the investigation of individual parameters [23].But the evaluation of reproducibility can only be carried out on the basis of repeated tests [67,80].From this, a test repetition of six times for each penetration depth results.To evaluate the reproducibility, minimising external parasitic influences has high priority.For this reason, all possible influencing variables such as temperature, cell bracing and the general cell condition (SOC and state of health (SOH)) are controlled (Sec.2.1 and 2.2).
The SOC of 100 % was deliberately chosen because it corresponds to the maximum cell reactivity [67,81,82].In addition, it should be mentioned that this high SOC is a probable initial state for an ISC caused by particle contamination [26] due to the increase in volume [49,83] or dendrite formation (e.g.Li-Plating [84]).Since the contact condition of the ISC cannot be identified from the voltage signal itself, it is meaningful to examine the most critical fault case here, the contact condition (Al↔An), in more detail.This is implemented by a specific cell geometry with a cathode as the outermost cell layer.
Preliminary studies have shown that ISC induces structural changes within the cell, which can lead to changes in the electrochemical behaviour [36,64,85].From this, the need to measure cell voltage and temperature (Sec.2.3) accurately is concluded.In addition, the force and the distance travelled by the needle as well as the gas emission are also measured to characterise the state of the ISC.This approach will be complemented by extensive post-characterisation, including the cycling of damaged cells (Sec.4.4) and a supplementary post-abuse analysis (Sec.4.5).Through this, extensive additional data on ISC behaviour will be recorded as well as evaluated, and statements on the risk assessment of ISC behaviour will be concluded.

Cells
The tested 10 Ah cell was specifically selected based on the following prior considerations: 1.
Creation of most critical ISC contact condition (Al↔An) by needle penetration requires Utilized cell chemistry (Graphite/NMC) is common for many applications and more safety-critical composition due to the low thermal stability with respect to alternative chemistries [1,86,87].

3.
Investigation of the known inhomogeneous failure characteristic of an ISC cannot be sensibly conducted at small (laboratory) cells [43,88,89] but with a cell capacity close to application values to generate transferable results.The corresponding ISC risk rises with the capacity due to lower cell-internal resistance R i and larger currents [29,30,90].
Based on the above requirements, a high energy NMC cell from Kokam, which was certified according to UL 1642, was used.This cell type consists of 23 double-coated anode layers, 22 cathode layers and two single-coated cathode layers as the outermost part of the electrode stack.The separator isolates the electrodes as a Z-winding and is made out of polyethylene (PE: melting point at ≈135 • C [91]).Further information of the cell can be found in Table 2.All cells originated from the same batch and were in new condition (100 % SOH) before the beginning of the experimental investigation.The tests were performed within 8 months and the cells were stored at room temperature in the delivered state (<30 % SOC) so that the calendar ageing can be neglected.Shortly before needle penetration -or other teststhe cells were preconditioned to check the proper cell condition and to reduce hysteresis effects [17].During this process, the cells completed three full cycles with 1C at 20 • C. In the last cycle, the cells were charged to 100 % SOC (CCCV until I < C/20).
The capacity measurement carried out identically on eight cells with 1C indicated a mean capacity of 9.994 Ah with a standard deviation of 0.023 Ah.This means that the relative standard deviation is only 0.23 %, which demonstrates a low spread of the cells.In addition, the literature reports a low capacity variation for this cell type [92] or in general for commercially produced LIBs [56].
In summary, the entire procedure ensured that the cells had a very similar initial condition, which is the basis for the most possible reproducible and comparable test results.

Surrounding cell conditions
Bracing [93] and tempering [94] have a major impact on cell performance already in the normal operating window of LIBs.This requires that the bracing and temperature must be precisely controlled during the abuse tests [95,96].For a realistic replication of the cell surroundings, the cells were tempered to 25 • C and braced with 0.11 MPa, which is inside the optimal operating range for lithium-ion pouch cells [94,97,98].
The cell was braced with six screws between two 8 mm thick aluminium (Al) plates.Between the Al-plate and the cell, a 3 mm thick insulation material made of glass fibre was placed (see Fig. 4 -Test Setup).To control the temperature of the cell, two heating cartridges were integrated into the bracing plate, which were controlled via a PID controller.For post-characterisation (Sec.3.3), the cell temperature was controlled by a temperature chamber.

Temperature measurement
For the temperature measurement, different K-type sheath thermocouples were used, and the temperatures were recorded via a data logger.It is known from the literature [43,55,88] as well as from our own preliminary tests that an internal short circuit leads to a very inhomogeneous heat distribution.For this reason, the cell temperature field was recorded by numerous temperature sensors (Fig. 2).The penetration point is marked by a dark brown cross on the pouch foil (reference contact).For local temperature recording near the penetration point, 0.5 mm thin sheath thermocouples (manufacturer: TC Direct; type: 406-613) are used.These temperature sensors can detect dynamic temperature changes due to their low heat capacity and are marked in red in Figure 2.With Kapton ® adhesive tape, the sensors are attached to the cell.Starting from the penetration point, three of these sensors are glued on both the right and left side in the horizontal x-direction to record the radial temperature distribution.These temperature measuring points (T − 9 * , T − 9 * * ,

T-Mean
For the measuring point T − Control marked in green the 1.5 mm sensor type is used as input value for the PID controller.For each test, a single-use medical cannula (manufacturer: B. Braun; type: Sterican ® Safety -21G x 1") was used as a penetration needle (see Fig. 3), which has a high sharpness that is identical for each individual needle [99].From this the best possible mechanical initial conditions for the creation of an ISC were derived.The material of the needle is chrome-nickel steel, which has a melting point above 1400 • C [100].This needle type has an outer diameter of 0.8 mm and an inner diameter of 0.57 mm, so that a sheath thermocouple (type K) can be integrated into the hollow needle to be able to measure the local ISC temperature as locally as possible [75,101].The thermocouple can move freely in the hollow needle and measures the temperature (T − Needle) at the cell surface.Due to the small distance between the two collectors (≈173 µm), the ISC is only caused by the foremost part of the needle (see detail in Fig. 3).

General experimental setup
For the precise needle penetration, a self-developed test rig was used, which is shown in Figure 5.This test rig consists of the upper part (orange), in which the actuators and sensors are installed, and the test chamber below (red), in which the cell is penetrated in a controlled procedure.The test chamber is connected to a fume extraction system in the rear section (green) (volume flow ≈200 m 3 h −1 ).To achieve a high mechanical stiffness, the test rig was constructed from robust aluminium profiles.Except on the one side where the air intake is located (blue), all other sides are closed with flame-retardant polycarbonate plates (Makrolon ® ).
The forward motion of the penetration needle is realised by the vertical movement of a spindle and is shown as a concept sketch in the upper area (Test setup) of Figure 4.A stepper motor is used as the drive.This motor is driven by a controller, which defines the set point position (X Setup ).The motor-driven spindle presses on a load cell.A precision sensor measures the position (X Measured ) of the spindle via a connecting plate with an accuracy of ±1 µm.On the bottom side of this plate, a shaft is mounted in the axial direction, which extends into the test chamber.A spring on the shaft provides a counterforce in the direction towards the motor so that gravity-induced lowering of the shaft is prevented.In the penetration area, a guiding sleeve for the needle is integrated into the bracing plate, thereby minimising electrolyte evaporation from the cell.An air quality sensor is installed in the test chamber, which can detect the typical released gases CO 2 , H 2 , TVOC (Total Volatile Organic Compounds) during venting of the cell [103,104].The TVOC concentration is particularly interesting as an indicator for electrolyte evaporation.To identify larger amounts of smoke or sparks and flames, the test chamber is filmed with a video camera.A data logger records the cell voltage and the voltage of the load cell's amplifier with 500 Hz sample rate.The comparatively high sampling rate is necessary due to the expected very dynamic voltage fluctuations [55].Before and after the needle penetration, the mass of the cell was measured with a scale to calculate the mass loss.All recorded data for the characterisation of the ISC are summarised in Table 3.

Method
The general process for the generation and characterisation of ISCs is illustrated as a flowchart in Figure 6.

Experimental Procedure
The generation and characterisation of the ISC was realised by a standard procedure, which is shown in individual steps in Figure 4 and described in more detail below.First, the cell is preconditioned and fully charged, as already described in Section 2.1.After a homogeneous temperature of 25 • C was ensured in the cell, the needle was inserted step by step ( 1 to 3 ) in the central position of Figure 4.In the first step 1 , the needle is controlled manually (step size 10 and 100 µm) towards the cell until an electrical contact is established    between the needle and a reference contact (see Fig. 4 in yellow).
This needle position is set to 0 µm for reference.Then, in step 2 , the needle is moved further into the cell with a speed of 1 µm s −1 until the first Al-Collector of the cathode is contacted.To generate the ISC, in step 3 the needle is moved deeper into the cell until the 1 st ISC is produced between the 1 st cathode layer and the 1 st anode active material (Al↔An).
A voltage drop of ∆U>1 mV is used as a criterion for the detection of the 1 st ISC.The actual force on the needle is referred to the state just before the 1 st ISC.At the time of the 1 st ISC, the time reference is set to t = 0 s and the penetration is either stopped immediately, which corresponds to an additional penetration depth of theoretically 0 µm (Case 1) or further driven into the cell for additional 25 or 100 µm to achieve a better contacting (reduction of R Al↔An ) between the needle and the anode (Case 2).Due to the individual ISC behaviour, six cells were tested at each penetration depth ∆X (0, 25 and 100 µm).Please find an overview of the test specifications and nomenclature in Table 4.
After triggering the ISC, the cell was observed until a clear ISC characteristic (Fig. 1) was formed (t < 5 h).If there was no TR of the cell, the needle was removed from the cell and the puncture was sealed with a fast drying epoxy resin, which prepared the cell for further postcharacterisation (see Sec. 3.3).To specify the exact initial contact conditions, a dummy cell was used, which was not electrochemically active due to the missing electrolyte.With this dummy cell, the typical fusing did not occur during needle penetration and thus there was no direct change in contact condition between the needle and the anode.The geometrical parameters of the dummy cell were identical to those for the further experiments.This dummy cell was penetrated four times until the first contact with the anode and four times until the contact of the second cathode, which corresponds to the complete penetration of the 1 st anode.To measure the contact resistance, a digital multimeter was used in a 4-wire configuration, and then the damage inside the cell was analysed in more detail using the digital microscope.The penetration depth at which the measured resistance value of the multimeter changes from infinity (∞) to a value in the kΩ range is defined as the 1st contact point.

Key features to describe the short term characteristics
The ISC is known for the highly dynamic, and for each ISC individual behaviour, as also illustrated in Figure 1.To compare the different ISC characteristics, several identifying features were defined, which are listed in Table 5.
The internal resistance R i and the open-circuit voltage U OCV of the cell are directly measurable or determinable cell-specific parameters and have a significant influence on the characteristics of the ISC.In the case of an ISC without any other external load, the voltage U ISC can be measured externally at the clamps.The resistance R ISC and the associated current I ISC , however, cannot be measured directly.Following Formula (2), it is possible to calculate the short-circuit resistance R ISC and, deduced from this, the internal short-circuit current I ISC and the short-circuit power P ISC as given by the Equations ( 3) and (4).
The short-circuit current (I ISC ) responsible for the voltage drops leads to a discharge of the cell.By continuously calculating the amount of converted electrical discharge, the actual open-circuit voltage U OCV can be calculated with reference to the open-circuit voltage curve stored as a lookup table.This makes it possible to calculate the corresponding short-circuit resistance, current or power for each recorded voltage value.Based on this, the most severe ISCs were determined and analysed for each individual short-circuit characteristic.This characteristic is summarised under the rubric ISC-Severity.In addition, the time after short-circuit initiation is given as t max, ISC and the corresponding SOC as SOC charact.ISC .
In addition, the general ISC characteristic can be described by further features.For this purpose, the temperature by T − 9 * max is used, which represents the maximum measured temperature at position T − 9 * in the specific cell test.The ISC behaviour is related to the corresponding ISC-Type (see Figure 1).At this moment, the corresponding time t Charact.ISC at which this type of characteristic can be clearly identified and the corresponding SOC of the cell SOC Charact.ISC is also listed.The mass loss related to the total mass of the cell was determined as m loss% and calculated based on Equation (5).Here, the index 0 and ISC represent the initial state and the state after ISC creation, respectively.

Post-characterisation
Many methods for the detection of cell-internal failure states are based on possible changes in the electrochemical cell behaviour [31,78], and therefore the post-characterisation of the damaged cells is an important basis for improved failure detection.In addition to the electrochemical post-characterisation (Sec.3.3.1),a post-abuse analysis was also carried out (Sec.3.3.2) in order to evaluate the degree of cell-internal damage and these results with the short-term characteristics as well as the electrochemical post-characterisation.

Electrochemical post-characterisation
According to the literature, various methods can be used for the more detailed electrochemical characterisation of distinct damaged cells [24,64,85].The techniques implemented in this study are briefly described below.All tests were carried out with a Bitrode battery test system (type: MCV 32-10-5).For comparison, a new, undamaged cell is always also tested as a reference.

Self-discharge
An existing ISC leads to an increased self-discharge of the cell [105,106].After the short-term characterisation, the cell voltage was measured for at least 4 days.The average self-discharge was then determined from the known open-circuit voltage characteristic of this cell type.If the cell reached the lower voltage limit (U < 2.7 V) within this period, the cell was charged CCCV to 50 % SOC and then the resulting current after 5 h was defined as the self-discharge current.

Incremental Capacity Analysis (ICA)
The ICA is known in particular from the ageing investigation of LIBs and enables conclusions about the characteristics of essential ageing phenomena such as the loss of active mass, loss of available lithium and the increase in internal resistance on the basis of the voltage characteristic [107,108].For this purpose, the differential/incremental capacity (IC) is formed according to Equation ( 6) by differentiating the charge quantity (Q) with respect to the voltage (U).
As a data basis, the cells are loaded with a small current of C/20 for two cycles to minimise cell-internal inhomogeneities.The second discharge curve is filtered first to smooth the voltage [109] and then the incremental capacity is calculated for each time step (1 s).

Pulse characterisation
Pulse characterisation is an established method to characterise the dynamic cell properties [107,110].In this study, cell states at 20, 50 and 80 % SOC were examined.After setting the SOC the cell was first discharged for 10 s with 1C and 50 s rest phase (I = 0 A) and then charged with the same current for another 10 s and subsequent rest phase.A common R-RC-RC cell model was fitted onto the pulse data-sets to acquire the parameters as f (SOC).

Cyclic stability
The further use of the damaged cells was simulated by cycling (Charging: CCCV; Discharging: CC).For this purpose, a selection of damaged cells was loaded over more than 175 full cycles.Here the charge and discharge current was 1 C (10 A).After every 25 full cycle, the self-discharge was determined by the already described five-hour CV phase at 50 % SOC (see paragraph Self-discharge).

Post-abuse analysis
The post-abuse analysis began with an assessment of the overall optical impression.Using a digital microscope the remaining width of each needle tip l Needle was measured and thus the melting process at the penetration needle was analysed (cf.Fig 3).Subsequently, the cells were discharged to 0 % SOC and disassembled according to [26].During disassembly, the damage characteristics of each cell layer were analysed by the following steps: 1.
photographing the considered layer 2. determining the melted separator area using a millimetre paper as a scale reference 3.
checking if collector foil is pierced (yes/no) This procedure is repeated for the next deeper cell layer until the first undamaged separator layer is reached.As a result, the number of melted separator layers is given as Layer Sep, Melt .
All layers together account to the total melted separator area, which is given by A Sep, Melt .

General short circuit development and characterisation
This chapter consists of the geometric analysis of the trigger conditions for ISC initiation (Sec.4.1.1)and the description and interpretation of the general ISC behaviour in Section 4.1.2.

Geometric trigger conditions
The penetration of the dummy showed that the force on the needle increases continuously, as expected with penetration depth.Only when the needle stopped penetrating was there a decrease in force, which resulted from the mechanical relaxation of the penetration device.As can be seen in Table 6, the tapered structure of the needle tip leads firstly with increasing penetration depth to an enlarging incision size, the greatest linear expansion of the incised area caused by the needle.Secondly, a decreasing incision size results for the deeper cell layers.
The damage image shown in Table 6 (state I) shows that the incision of the separator closes itself again after the needle has been moved out (see Tab 6a.), as a result of which no clearly visible incision area remains.In this condition 3 , the incision size is only ≈27.5 µm on average, which means that the contact area for the 1 st ISC is very small.Due to the surface load acting in the puncture area, the separator structure is irreversibly compressed, with the effect that the separator becomes more transparent [111].The needle transports graphite into the puncture area, which becomes visible through the dark colouring [63].On the anode side (Tab.6c.), a recess with an average diameter of 75 µm and a depth of about 50 µm is formed in the graphite.From this increasing diameter it can be deduced that the indentation is primarily caused by the prevailing pressure effect and only to a limited extent by the needle penetration.Through the penetration, the needle tip enlarges from about 10 to ≈40 µm.The deeper penetration through the anode (state II) leads to significantly more damage, with an average incision size of 202 µm at the separator (Tab.6b.) and 212 µm at the anode (Tab.6d.).After penetration, the needle tip has an average width (l Needle ) of 45 µm.In more than 75 % of the tests, however, smaller distances were necessary for triggering the 1 st ISC.Taking into account the additional layer compression, contact with the Cu-Collector can be assumed to be very unlikely.These results are in agreement with the test results that were carried out with the dummy set-up.It can therefore be safely expected that the most critical short-circuit case, the contact between the Al-Collector of the cathode and the graphite of the anode (Al↔An), is generated by this needle penetration.On the basis of a total number of 26 tests (18 tests + 8 preliminary tests), it was found that after the very first contact with the 1 st Al-Collector, the needle still had to move an average of additional 167.52 µm into the cell until the 1 st ISC was formed.The standard deviation σ is 48.94 µm.It can be seen that often a longer distance X than the theoretical distance of at least 106 µm had to be covered (see Fig. 4).This behaviour can be explained Table 6.Penetration tests performed with the dummy cell: Exemplary illustration of the failure image for the 1 st separator and the 1 st anode in each case after the 1 st contact with the anode (penetration depth 3 ) and after the complete penetration of the anode, which corresponds to the first contact of the 2 nd cathode.by the local compression of the cell caused by the needle, so that the needle has to travel a longer distance.Based on this behaviour, the individual spread can also be explained, whereby for this concept the achieved standard deviation is still considered to be satisfactory.Theoretically, a distance of at least 193 µm is required for contacting the Cu-Collector of the 1 st anode.

General characteristic
The first voltage drop of the 1 st ISC is on average 6.1 mV by considering all 18 cells, so that according to [34] a minimal ISC (U < 0.05 V) was initiated by the needle at first.If there was no further needle penetration (∆X = 0 µm) after the occurrence of the 1 st ISC, the ISC could be finished within a few seconds, and consequently the cell showed a normal behaviour identical to an undamaged cell.In this case, the fusing at the contact point was so fast that the heat input at the ISC position was very low.It follows from this that no change in temperature and in the active force F on the needle was measurable.Yet typically, dynamic ISC behaviour (type II) occurred, which generally poses a greater safety risk.In the following, the generally valid correlations are described on the basis of exemplary test results.Here, cell 1.6 is examined more closely, since this cell reached the highest temperature T − 9 * at a penetration depth of 0 µm (Case 1) and accordingly showed increased internal cell damage (See Sec.4.5).Subsequently the characteristic ISC key features as described in Sec.3.2 are applied to cell 1.6.
Figure 7a shows the typical ISC characteristics for the period of the first 5 s after triggering.Immediately with the triggering of the 1 st ISC, there is a significant decrease in the mea-sured force (orange).After about 2.4 s, the needle penetration stops, which is indicated by the needle travel shown by a green dotted line.From this point on, no deeper penetration is initiated and the travel distance X Setup remains constant until the needle is removed.The lower part of Figure 7a shows the temperature curves of various sensors (see Fig. 2. Immediately after the occurrence of the 1 st ISC, an increase in the needle temperature can be detected.With a slight delay (≈3 s), the temperature sensor T − 9 * also detects a temperature rise.In the next Figure 7b, a longer time period of 650 s is shown for the same cell 1.6.

Correlation of voltage drop, joule heat and temperature
It is clearly visible that the cell voltage U drops more and more frequently and a dynamic voltage behaviour develops.The intensity of the voltage drops is easier to classify with the help of the calculated open-circuit voltage characteristic (light blue line).In addition, the local joule heat P ISC of the ISC is shown in grey in the lower part of Figure 7b.Through the additional observation of the temperature curves, the relationship between released joule heat and the direct temperature increase is shown.It can be seen that the needle temperature (T − Needle) shows the highest temperature and directly follows the dynamics of the released heat output.In particular, the more severe ISC, about 570 s after short-circuit initiation, shows this correlation very clearly.Here, the electric heat power was about 50 W and was less than the maximal power (74 W) of the cell.The extreme localisation of the ISC is shown by the measurement data of the other temperature sensors.An ISC leads to a temperature increase with clear temperature gradients in the radial direction around the penetration position.At time t = 600 s the temperature T − 9 * (0.7 mm distance from ISC position) was about 165 • C.However, the sensor T − 9 * * , which was only 0.7 mm further away from the penetration position, only measured 80 • C. Compared to these sensors (T − 9 * and T − 9 * * ), the sensors T − 9 ′ and T − 9 ′′ , which were located at the same distance from the ISC, measured very similar temperatures.The average cell temperature (T − Mean), however, only increased for 1.5 • C.

Correlation of temperature and local structural damage
With additional local heat input, structural changes occurred at the puncture position and consequently the needle continued to "burn free".If the intensity of heat output and resulting temperature is compared with the progression of the force, it can be clearly identified that the decrease in force occurs only after new maximum temperature levels have been reached (see Fig. 7a).In the case considered here in Figure 7b (cell 1.6), this behaviour can be observed after 23, 54, 240 and 372 s.However, the effects of the local force measurement become less and less significant.Besides, the travel distance sensor can also detect a change in travel X Measured , which occurred with a slight time delay to the measured decrease in force.The existing pre-tension on the needle led to a mechanical relaxation due to the reduction of the force.Minimal measured travel changes were the result.For clarity, the data recorded by the gas sensor are not shown in Figure 7b.In general, slightly increased gas concentrations could be detected above a needle temperature of 80 • C in all experiments.However, significant concentrations occurred only above a temperature of about 200 • C measured by the T − 9 * sensor.At first, the local needle penetration led to only a minimal outgassing of electrolyte.

Overall characteristic
Based on the entire ISC progression, shown in Figure 7c, the overall characteristic can be identified.The most severe ISC occurred with a calculated ISC resistance R min, ISC of 0.113 Ω after 321 s of ISC development.This corresponds to a current of I max, ISC 37.7 A and a joule heat P max, ISC of 143 W. Since these high ISC currents flow only for a very short period of time, the calculated SOC at this time (t = 321 s) was still 99.43 %. Figure 7c clearly shows that the cell does not lead to a TR (t = 600 to 900 s) even at very high   local temperatures (T − Needle > 1260 • C) and also keeps a high current dynamic.This behaviour was caused by the fact that the ISC is only very locally effective and, in contrast to homogeneous cell heating or an external short circuit, there is no kinetic limitation over a large area in the cell [30,112,113].A high local temperature is therefore not a reliable criterion for a critical cell state leading to TR.Shortly before a more constant ISC was formed as a result of an abrupt fusing after 992 s at a SOC of 84.92 %, T − 9 * max reached 314 • C.This was followed by a much more constant short-circuit behaviour according to type IIa, which was characterised by continuous cell discharge and lower temperature dynamic.The mass loss m loss% of this cell 1.6 was 3.5 %.

Influence of penetration depth
With deeper needle penetration, a better contact between the needle and graphite was generated, which led to a decreasing resistance R ISC (see Fig. 4).The resulting ISC behaviour at different penetration levels (0, 25 and 100 µm) is analysed in this chapter based on the previously introduced characteristic features.Cell behaviour of particular relevance will be discussed here in more detail.

Penetration depth of 0 µm
Table 7 lists the key features according to Tab. 5 for six cells in which penetration stopped immediately upon ISC entry (penetration depth ∆X = 0 µm).In addition, the specific cell behaviour is described by the following key points.

•
In experiments 1.2 -1.6 there was a dynamic development of the ISC (type IIa), in which the most severe ISC occured within 116 s (cell 1.2) and 337 s (cell 1.3) after the 1 st ISC.
• The maximum internal short-circuit currents of cells 1.2 -1.6 were between 11 A and 38 A.
• Local temperatures (T − 9 * max ) above the decomposition temperature (T Onset = 250 • C) of the cell components did not necessarily lead to a TR (cell 1.6).The measured temperatures at the needle were typically above 500 • C.

•
The characteristic formation of a constant ISC for type IIa occurred with a clear delay (567 to 992 s) in the high SOC (85 to 97 %).
• Corresponding to the highest temperatures, a mass loss of 0.3 % was measurable only for cell 1.Consequently, it can be stated that in the case of minimal contact with the active material (penetration depth ∆X = 0 µm), as is for example typical for a dendrite, there is no immediate safety-critical condition.However, due to the remaining cell-internal structural weakening (see Sec. 4.5 -cell 1.6), these cells pose a risk for further use that is difficult to calculate.All results presented in Table 8 show a dynamic ISC behaviour (type II).However, each experiment showed an individual characteristic as described by the following points:

•
In the experiments of cells 2.1 -2.5, the most severe ISC with a joule heat (P max, ISC ) from 48 to 80 W was in the same range.However, the time of occurrence was very different (cf.t = 9 s for cell 2.3 and t = 650 s for cell 2.4).
• The characteristics of cell 2.1 -2.5 correspond to type IIa, which only appeared with a significant time delay (500 to 965 s).
• Cell 2.6 showed an abrupt drop of the voltage below 3 V after 363 s.The released maximum electric heat power P max, ISC ) was 884 W. At this point, the cell was still almost fully charged, with over 98 % SOC.The heat released by the powerful ISC initiated exothermic side reactions and caused a TR.Just 9 s later, the cell voltage dropped to 0 V (type IIb).
Based on the key features, the ISC of cell 2.4 can be classified as less severe compared to cell 2.5, as shown e.g. by the maximum temperature T − 9 * max .
The special behaviour of cell 2.4 is the first dynamic short-circuit development, which came to an almost complete stop after 28 s and resulted in a nearly constant voltage.About 500 s later, dynamic ISC behaviour developed again, which is shown in Figure 8.Further 100 s later (650 s after the 1 st ISC), a stronger ISC occurred for nearly 2 s, and it released a joule heat of about 40 W locally in the cell.The mean cell temperature (T − Mean) raised by 17 • C within 15 s and for a short time the rate of temperature increase was more than 1 • C s −1 , which is often taken as a threshold to indicate a TR [114,115].In addition, an abrupt swelling of the cell as well as a clear release of smoke gas was visible in the video recording.The CO 2 concentration measured by the gas sensor jumped within one second (corresponds to one measured value) from the reference value of 412 ppm to the maximum possible measured value of 57 330 ppm, which confirms the significant gas release.Such a characteristic corresponds for the most part to the behaviour of an initiating TR.However, compared to a typical TR, cell 2.4 showed an early stop of the temperature rise and the voltage drop.This behaviour, here named as Partial Thermal Runaway is not known from previous literature and shows that new relevant results can be generated with this precise needle penetration.
If the behaviour of the cell 2.5 is considered, it shows a comparatively slow temperature increase with a maximum temperature increase of the average cell temperature of 8 • C within 100 s.Here the maximum gas release was only 1647 ppm although the local temperatures were increased.The reason for the more intense reaction of cell 2.4 was the triggering of exothermic side reactions, which led to an increased heat release and the mentioned gas generation.Out of a total of 26 needle penetrations, only one other cell (cell number To summarise, neither local temperatures nor specific voltage drops can be used as a generally valid criterion for safety assessment.It seems more likely that the initiation of exothermic side reactions is related to further complex criteria, in particular the previous ISC development.The TR itself can occur without prior typical characteristics, and therefore especially at high SOC, the occurrence of a dynamic ISC must already be classified as generally safety-critical.The unusual behaviour of the cell 2.4 underlines the difficulty of characterising the ISC behaviour by a limited number of key features.

Penetration depth of 100 µm
Due to the melting process (fusing) at the needle tip (>30 µm), it can be ensured that there was no contact with the Cu-Collector even with an additional penetration depth of 100 µm.This even deeper penetration of the needle led on average to a stronger ISC behaviour than the previously presented test series (1.X and 2.X) with lower penetration depths show.The following characteristics can be concluded from the results presented in Table 9:

•
In four out of six tested cells, a TR developed from the dynamic ISC (type IIb).Here, the joule heat (P max, ISC ) of the ISC was more than 800 W.
• Many cell components combust in the TR, and consequently the cell mass was reduced by 52 to 58 %.The maximum temperature (T − 9 * max ) was between 540 to 607 • C.
• For cells 3.1 and 3.2, a constant ISC (type IIa) developed from the dynamic ISC behaviour.This process occurred with a significant delay of more than 36 min for cell 3.1.
The results presented in Table 9 show the typical scatter.However, it is noticeable that the most severe ISC of cell 3.3 -3.6 shows a similar characteristic, which led to the TR of the cell.For local cell heating, not only the heat power but also the amount of local heat energy is relevant.Figure 9 shows the voltage characteristics and the calculated joule heat for two  In the time period from 0.2 to 0.35 s, there is a strong relaxation of the voltage.This is followed by another significant voltage drop.Within a few milliseconds, the cell voltage rises again -probably as a result of fusing.During the parallel TR, the cell burned out layer by layer.When the last cell layer has burned after ≈8 to 12 s, the cell voltage dropped to 0 V and the TR is completed.At this time, the mean temperature of the cell reached its maximum value.The cells 3.4 and 3.5, which are not shown here, also had a similar behaviour.Furthermore, cell 2.6 (∆X = 25 µm), which runs also into the TR, corresponds to this specific behaviour.In the case of cell 3.6, a significant internal short-circuit already occurred 90 s before the severe voltage dropped.The result was an increased charge max .From the voltage drop (∆t = 0 to 0.4 s) shown in Figure 9, it can be seen that the short-circuit resistance decreases comparatively slowly until the relaxation of the voltage occurs after about 0.42 s.About 10 s later, here the voltage also drops to 0 V as a result of the TR.Despite the different progression of the voltage drop, in both experiments (3.3 and 3.6) there is a clear relaxation of the voltage after ≈0.6 s.When the mean power is calculated for this period (0.6 s), this results in about 321 W for cell 3.3 and 500 W for the cell 3.6.

Summary and interpretation of the short-term ISC behaviour
The initiation of the 1 st ISC is subject to statistical scatter, although intensive efforts were realised to ensure a reproducible experiments.Due to the individual contact resistance R Ca↔An , a wide variety of initial conditions result in the further development of the internal short-circuit behaviour.However, these individual contacting conditions are in principle also expected for every ISC occurring in practical application, and as a result a direct correlation exists between the replicated and real ISC.
From the totality of all results, an extended, new model approach can be derived, which is shown in Figure 10.Based on the dummy experiments, it is known that the resistance R Al↔Ip between the needle, here generally referred as impurity particle (Ip), and the Al-Collector is much smaller than the contacting resistance R Ip↔An between the needle tip and the graphite anode.Thus, the ISC characteristic is essentially determined by R Ip↔An .The resistance R Ip↔An can be simulated by a resistor network, as sketched in Figure 10.The observed short-term fluctuation is indicated by the switchable resistors to create the sharp voltage drops.In addition, the long-term ISC development is emulated by the controllable resistors themselves.
Local fusing can interrupt one or more current paths, which is symbolised in Figure 10 by the grey path with an open switch.The heat input from the ISC causes various subsequent processes, such as the initiation of exothermic side reactions [64], electrolyte evaporation [65] and the melting of the separator [22,23], as a result of which the local resistances change dynamically.
In the tested high SOC (100 %), the required activation energy for the initiation of exothermic reactions is reduced and a dynamic, difficult-to-predict ISC development occurs.If the needle is inserted deeper into the anode, a lower short-circuit resistance results.The reason for this is, that on the one hand, there is a larger contact surface at the tip of the needle, and on the other hand, the distance to the copper collector is reduced.It is striking that in five out of six cases the TR develops spontaneously from a previously very moderate, local temperature increase (T − 9 * < 60 • C).On the other hand, high local temperatures (T − 9 * > 160 • C), do not induce TR of the cell.During dynamic ISC development, local discharge occurs at the short-circuit area [116], which increases the required activation energy to initiate exothermic side reactions [67,82].The longer the dynamic discharge continues, the greater the required energy input to initiate the TR.A slow local temperature increase leads to the evaporation of the electrolyte, which contributes to further passivation at the short-circuit area by acting as a missing reactant.Furthermore, it is known that a slower heating results in a reduced maximum heat release of the running exothermic side reactions [117,118].Due to these processes, local temperatures of over 1260 • C were repeatedly measured at the needle in various experiments, without any accelerated, large-scale progression of exothermic side reactions.The individual reaction condition and process at the short-circuit location cannot be determined from the outside, which complicates the ISC assessment.
Neither the local temperature nor the cell voltage is any criteria for a generally valid risk assessment.Instead, dynamic ISC behaviour itself must be classified as potentially dangerous.Additionally, a further risk assessment based on the known SOC is meaningful.Compared to the external short circuit or the slow heating of the cell, there is no largescale pore closing of the separator in the case of ISC due to the high locality and thus the entire cell is not limited by kinetics.Due to the very inhomogeneous temperature distribution, temperature measurement is not a suitable parameter for failure detection.The entirety of the tests showed that the initiation of a single-layer ISC type (Al↔An) with only a small contact area of the needle can trigger a TR of the cell.As a result, even the smallest production-related defects or contamination with impurity particles [26] must be considered as safety-critical.

Electrochemical post-characterisation
The electrochemical post-characterisation is an important complement to the ISC characterisation.An overview of the cells considered in more detail is shown in Table 10.

Self-discharge
Only a few conclusions regarding the previous short-circuit behaviour can be deduced from the measured self-discharge.Cells with a minimal ISC behaviour (type I) did not show an increased self-discharge.In case when the cells shows a dynamic ISC behaviour (type II), very different characteristics can occur, as the results from Table 11 show.The cell 1.6 reached the highest local temperature of 314 • C, but the self-discharge was only about 19 mA.Compared to this, cell 2.1 showed an increased self-discharge despite lower previous local temperatures (T − 9 * = 82 • C).Initially, the self-discharge of cell 2.1 was about 800 mA and dropped to roughly 400 mA within a few days.The cells 2.4 and 0.1 showed an abrupt temperature increase in the ISC behaviour (Partial Thermal Runaway), which can be explained by the progression of exothermic side reactions (see Figure 8).Nevertheless, both cells have only a slightly increased self-discharge.Within seven days, the voltage of cell 2.4 was reduced from 4.12 only 4.042 V, which corresponds to a self-discharge of about 4 mA.The largest self-discharge was measured at cell Table 11.Results of the post analysis for six cells damaged by needle-penetration caused ISC (SOC = 100 %) with respect to the reference cell.Characteristic features of the non-destructive (self-discharge) and destructive analysis.For reference, the ISC behaviour is given (ISC-Type and T − 9 * max ).

3.1.
Here it was noticeable that even after five days, the internal short-circuit current still showed a high dynamic.After about 4.5 d, the current increased from 1.1 to 1.7 A within a few seconds.This observed behaviour causes an incalculable safety risk and can be a plausible cause for the time-delayed TR observed in practice after a damage event [5,119].

Electrochemical behaviour via ICA (and Pulse)
The pulse characterisation of the damaged cells did not show clear changes compared to the reference cell and is not shown here.This behaviour can be explained by the minimal area-related damage to the cells.The only locally effective temperature input means that the majority of the cell remains undamaged.For all cells, the sometimes increased selfdischarge is much smaller than the pulse load, and therefore there is no relevant influence on the short-term characteristics.This behaviour corresponds to the result shown in [64], where no change in the short-term characteristics of damaged cells could be measured by electrochemical impedance spectroscopy (EIS).
The incremental capacity is shown in Figure 11 for three damaged (1.6, 2.4 and 0.1) cells and the reference, which corresponds to cell individual 1.6 in the new state.Cell 1.6 shows the same characteristic after the penetration, which is visible by the overlying green and orange curves.The red curve of cell 2.4 shows a slightly right-shifted peak, which can be an indication of an increased internal resistance [120].However, even the largest first peak at 3.65 V is not significantly reduced, which indicates a small loss of active mass.For cell 0.1 the black curve indicates a general reduction of the peaks, which can be related to a reduced active mass.Especially the 3 rd peak at 3.47 V shows a significant reduction compared to the reference.This behaviour is typical if the cathode loses active mass in the delithiated state [120], which is a plausible scenario due to the high SOC (100 %) of the tested cell.In terms of capacity, the anode of commercial cells is oversized [121].Consequently, this decrease in the capacity of cells 2.4 and 0.1 at a low current load (here: C/20) can be plausibly explained by the loss of active mass of the cathode.In general, it must be pointed out that the changes in the incremental capacitance are small.Cell 1.6 considered here, with the highest local temperature impact, also shows a normal, inconspicuous behaviour in this respect.Hence, the small changes in the electrochemical behaviour cannot be used as an indication of the internal extent of damage or the failure history.

Cycle stability
The results for a reference cell (green) and three damaged cells (1.6, 2.1 and 0.1) are shown in Figure 12.At the start of cycling, the reference cell had a discharge capacity of about 10.2 Ah.During the first 125 cycles, there was a slight increase in capacity of about  0.1 Ah, which is typical for LIBs [121].Due to the moderate load, no ageing of the reference cell could be detected, and self-discharge stayed in a low level over the entire test period.Cell 1.6 showed an almost constant high capacity over 175 cycles.The self-discharge of 60 to 80 mA was slightly higher than the reference cell, but also quite constant.In the case of cells 2.1 and 0.1, the previous stress caused by the internal short circuit resulted in a notable decrease in capacity to about 9.6 Ah.Increased self-discharge of cell 2.1 (≈400 mA) led to a reduced capacity of 0.4 Ah during the discharge of about one hour.A large part of the capacity reduction was consequently caused by the increased self-discharge and only partly by typical ageing phenomena.As a result of the cycling of 350 full cycles in total, the self-discharge was reduced to about 200 mA over time.In contrast, cell 0.1 showed a lower self-discharge of less than 50 mA.The decrease in the capacity of cell 0.1 after needle penetration can therefore be attributed to an internal cell degradation, which can also be seen in the incremental capacity analysis (see Fig. 11).After 25 cycles, a slight increase in capacity was measurable.Subsequently, the cell capacity was almost constant for another 225 cycles.However, after 255 cycles, the cell 0.1 suddenly suffered TR, which is illustrated by Figure 13.At this time, cell 0.1 was in the CV phase of the charging process and the charging current had already been reduced to 0.2 A. Without any previous abnormalities, a voltage drop of 10 mV occurred, which is defined as the reference time 0 s here.Due to the set point specification of the CV phase (U cell = 4.2 V), the current of the test bench was increased so that the voltage drop caused by the internal short circuit could be compensated first.
Yet, after 7 s, the ISC became so powerful that the maximum charging current deliverable by the test bench of 10 A was reached.At this point, no increase in cell temperature could be measured.Just one second later (t = 8 s), the temperature T − 9 * directly increased to 53 • C. A short time later, the TR of the cell occurred.After 18 s, the cell voltage dropped to 0.4 V and the cell was completely burnt out.The test bench continued to supply 10 A, which flowed across the remaining contact of the collector foils.As a result, a dynamic voltage characteristic followed, which is visible for the rest of the record (Figure 13-left).The abrupt voltage drop corresponds to type IIc and was thus significantly more severe than the voltage drops previously caused by the needle.Rather, the failure progression here corresponds to that of a nail penetration [69,122].On the basis of the behaviour of this cell 0.1, it is clear that cell-internal damage can lead to spontaneous TR during cycling.Due to the presented rapid failure development (t<20 s), the result is that such a failure case can not be detected early enough.In this case, it is more promising to detect the failure when the 1 st ISC occurs in order to stop the further use (cycling) of such a defective cell.

Post-abuse analysis
The post-abuse analysis shows that there are plausible correlations between the degree of optical damage, the maximum reached temperatures and the melting behaviour of the needle.Those cells 2.4 and 0.1 with a Partial Thermal Runaway, showed a clearly divergent damage structure, which can be related to the additional heat input.Table 11 lists the visual damage (++ minor damage to −− severe damage) and the other results of the post-abuse analysis.
Cell 1.1 showed only a slight local temperature increase, and consequently only a minimal puncture could be seen on the cell surface.The minimal ISC (type I) resulted that the melt area of the separator (A Sep, melt. ) had a diameter smaller than 200 µm.Also, the remaining width of the needle tip (l Needle ) of 110 µm was only slightly increased compared to the new needle (cf. Figure 3), which corresponds to a low shortening of the needle length.In the case of a dynamic ISC behaviour (type II) which occurred in the other cells, the higher heat input led to increased cell-internal damage.Of all the cells in which no TR occurred, cell 1.6 had locally heated up to the highest temperature of 314 • C at T − 9 * .This resulted in a large area of melted separator (A Sep, melt.= 36.3cm 2 ) and a large agglomeration at the needle tip.Down to the 5 th separator layer (Layer Sep, melt.), local melting of the separator can be detected here.The development of exothermic side reactions resulted in a clearly different damage structure in cell 2.4.Here, a large-area (A Sep, Melt = 196.8cm 2 ) but less deeply immersive melting behaviour (Layer Sep, melt.= 3) of the separator could be identified.In the following, the internal damage of cell 1.6 is discussed first, followed by general correlations.Based on these findings, the internal damage of cell 2.4 is then described.Finally, the damage of a burnt-out cell (TR) is shown.

Cell 1.6
The increased local temperature triggered the gas release at the penetration point.This resulted in bumps, which are visible in pictures A and B of Figure 14.On the detailed image of the 1 st cathode (Fig. 14.C), a melted area of approx.3.6 mm is recognisable and the active material of the cathode has separated from the Al-Collector on an area of approx.3 cm 2 .As a consequence, there is an increased risk of a more critical short circuit with direct contact to the Al-Collector (Al↔An).The neighbouring separator (Fig. 14.D) shows a round melting area around the penetration point, which will be discussed in more detail later (see Fig. 15a).Compared to the cathode, the anode (Fig. 14.E) shows an enlarged melting area of the Cu-Collector (∅ = 4.3 mm).A similar damage structure was also observed by Y. Yokoshima et al. and can be explained by the high local joule heat [89].
Copper has a melting temperature more than 400 • C higher than aluminium, and therefore the heat hotspot, which caused the heating of deeper cell layers, can be allocated to the anode.Figure 14.F shows the next 2 nd cathode, on which residues of the anode active material were found.However, the associated Al-Collector was not pierced, thus a contact between the two active materials (Ca↔An) was formed in the 2 nd to 5 th cell layer.With a deeper cell layer and therefore a greater distance to the puncture point, the melting area of the separator reduces (Fig 14 .G).The Figure 14.N shows the needle, which has a dark clump of about 2 mm at the tip.
In Figure 15a, the melting surface of the 1 st separator layer is measured and the maximum temperatures are plotted (see temperature curves in Fig. 7c).It can be seen that the high temperatures (T − 9 * * and T − 9 ′′ ) lead to complete melting of the separator.At the position of T − 9 * * * the maximum temperature was 138 • C and a transparent area of the separator is visible.In contrast, the sensor T − 9 ′′′ only reached a maximum of 117 • C, which did not lead to any melting of the separator.
From this context, it becomes clear that in general the maximum internal reached temperature can be reconstructed from the melting behaviour of the separator.In the practical analysis of damaged cells with unknown state, this correlation can help to evaluate the internal state quantitatively.The melting area A Sep, melt.for all damaged individual separator layers is plotted in Figure 15b.Here, layer 0 corresponds to the external insulation (see picture B in Fig. 14) and shows a lower melting area than does layer 1, where the internal short circuit was initiated.For the entire time of the dynamic ISC, this temperature hotspot remained at the insertion point.However, no significant thermal energy seems to be released by this ISC contact condition (Ca↔An) itself, which was formed in the deeper layers.For the complete cell, the melted area was 36.3 cm 2 .In the results of the electrochemical post-characterisation Sec.4.4, this cell 1.6 showed only minor abnormalities, with a slightly increased self-discharge (I Self = 19 mA) and a minimal decrease in capacity.This behaviour can be explained by the high electrical resistance of both active materials (σ ≤ 1 S cm −1 [123]), with the result that even a large-area contact only leads to a low ISC current.Furthermore, non-conductive separator residues remain in the short-circuit area, which additionally reduce the electrical contact.

General correlations
When the thermally induced damage characteristics were collected for all cells, excluding cells 2.4 and 0.1, the new relationships shown in Figure 16 were the result.The greater the maximum temperature T − 9 * max was, the greater was the separator melting area A Sep, melt. .Furthermore, it is clear from Figure 16a that the number of damaged separator layers Layer Sep, melt.also increases with a higher maximum temperature.Figure 16b shows  Cell 2.4, as well as cell 0.1, showed a clearly divergent behaviour inform of a Partial Thermal Runaway. Figure 17 shows the corresponding damage characteristics.In Figure 17A, the pouch foil is clearly damaged as a result of the sudden gas release.Furthermore, a wrinkled structure of the 1 st cathode's Al-Collector and an almost completely melted isolation are visible in Figure 17B.The further Figures 17D and E clearly show the pronounced decomposition of the anode.It is noticeable that the graphite of the anode was detached from the collector to a much greater extent than the NMC of the cathode (Fig. 17C).The detached active material increases the risk for a more critical ISC type with direct contact to the collector in the case of further electrical or mechanical loads.If a massive exothermic reaction of the cathode occurs, the heat of reaction typically induces complete melting of the Al-Collector [70,124].The missing melting process here can be explained by the thicker, thermally more stable 1 st collector foil (see also Fig. 18).In Figure 17E, the needle penetration into the anode is clearly visible, whereas the deeper cell layers do not show any penetration of the needle.The 3 rd separator layer shown in Figure 17G is melted in the centre, but in a non-uniform structure, and the following layer (Fig. 17H) is already undamaged.Due to the thermally insulating effect of the cathode active material and separator as well as the heat capacity of all components, the short-term released heat input from the decomposition of the anode did not reach so deep into further cell layers.Figure 17N shows the penetration needle with a clear melting area at the tip of the needle.
The extensive internal damage to the 1 st and 2 nd layer and especially the visible structural changes at the 1 st anode suggest that the original cell capacity is reduced by at least the amount of one layer.However, due to the over-dimensioning of the anode, the decomposition of the cathode in particular was responsible for a reduction in the cell capacity.In the new state, the cell capacity measured with a current of C/20 was 10.92 Ah and in the damaged state it was still 10.53 Ah.Thus, the capacity was only reduced by about 0.39 Ah.A single-sided cathode coated with active material has a nominal capacity of about 0.23 Ah.Accordingly, it can be assumed that the 1 st cathode (Fig. 17C) and also partly the 2 nd cathode (Fig. 17F) decomposed thermally.

Cell with TR
All cells that ran into a TR show a similar damage characteristic, which is depicted in Figure 18 by the representative cell 2.6.Due to the high heat input, the insulation material melted together with the pouch foil, as indicated in Figure 18a.The remaining combustion deposits of the active materials were no longer adherent to the collector.In Figure 18b.and c. a hole of more than 10 mm could be detected at the area of the puncture point.In this case, the 1 st Al-Collector also remained largely intact, because of the enhanced thickness of the material (Fig. 18b).The TR caused the internal Al-Collector of the cathode to completely melt, and thus only aluminium droplets could be found in the combustion deposits.Especially from the structural change of the 1 st Cu-Collector (Fig. 18c.), the gas release starting from the penetration point becomes very clear.These structures that are oriented towards the centre occur up to the 4 th Cu-Collector (Fig. 18d.).A needle puncture can be seen up to the 7 th Cu-Collector (Fig. 18e.). Figure .18f. shows the 8 th Cu-Collector, which has a shapeless, but not penetrated structure.
Literature has already shown that local gas release occurs during the ISC, which then causes bending apart from the surrounding cell layers [71].The resulting bulge in the direction of the penetration needle represents a low mechanical resistance, and consequently the next deeper cell layer was penetrated.From the damage (see Fig. 18), it can be deduced that the initiated ISC triggers the exothermic decomposition on the first 4 layers.This resulted to cell-internal reaction propagation over all layers.Due to the increase in volume of the entire cell stack, there is mechanical penetration of the already decomposed cell layers 5 to 7. In the deeper cell layers, neither a local fusing nor an outgoing gas release is visible at the puncture site (see Fig. 18e.).
The needle shown in Figure 18g.comparatively, was less molten of about 0.4 mm.From the post-abuse analysis of all cells with TR behaviour, a very similar damage characteristic was found.This is characterised by the fact that always the first 3 to 4 Cu-Collector layers have the structure similar to Figure 18d.and oriented towards the centre with a clear melting area at the puncture point.The deepest puncture always occurs in layer 5 to 7. In contrast, cells 2.4 and 0.1, in which the process of exothermic side reactions was limited to the 1 st cell layer, show only a single-layer needle puncture.It can be deduced from this that the gas release of cells 2.4 and 0.1 does not induce the penetration of the 2 nd cell layer, and thus the progression of the reaction was stopped.Consequently, it can be stated that under these ISC conditions, only a multi-layer ISC leads to the complete TR of this 10 Ah cell.

Conclusions
The replication of a local, single-layer ISC, as caused by contamination with impurity particles or when dendrites grow, is especially relevant.It is precisely this failure condition that can be reliably reproduced in pouch cells through the newly developed method in this work.Therefore, a sharp, thin medical cannula is penetrated in the cell with a speed of 1 µm/s.In this way, a reproducible mechanical damage scenario is generated in the cell, which is a great advantage compared to currently known trigger methods.For the cells used here (cell: 10 Ah; Graphite/NMC) with an outlying cathode, this generated a particularly critical contact situation between the Al-Collector of the cathode and the graphite active material of the anode.A defined tempering (25 • C) and a constant tension (0.11 MPa) were used to simulate realistic cell conditions and minimise any external influences.The occurring individual ISC characteristics can therefore be attributed to the scattering contacting resistance of the ISC.
The result is that at 100 % SOC, dynamic short-circuit development (Type II) is likely even at a very small penetration depth (<5 µm).Deeper penetration results in better contacting, which makes spontaneous thermal runaway of the cell more likely, such that in practice even small impurity particles ≈100 µm must be classified as very dangerous.The typical fusing and the associated local structural changes can be characterised by the change in the locally effective force on the needle.A local temperature measurement shows the direct correlation between the voltage drop of the cell and the immediate temperature increase.The ISC is characterised by a very local temperature increase, which initially shows the cell to be undamaged from the outside.A local temperature of over 1260 • C, which causes large-scale, multi-layer melting of the separator, did not lead to thermal runaway of the cell.Rather, a critical state develops after a comparatively mild failure characteristic, if a high SOC is locally present and a short-term high joule heat input occurs at the same time.The post-abuse analysis showed that no TR occurs for the tested cell type as long as a multi-layer ISC does not develop as a result of local gas release.In many cells, especially at penetration depths of 0 and 25 µm, local joule heat release mainly occurred and no large-scale process of exothermic side reactions followed.It can be seen that even a largearea contact between the two active materials (Ca↔An) does not lead to an immediate safety-critical state of the cell.
In two cells, the first cell layer suddenly decomposed without initiating further exothermic reactions in deeper cell layers.To our best knowledge, such a behaviour, which is named as a Partial Thermal Runaway has not been documented in the literature yet, and it underlines the challenge in developing a uniform definition of the TR.
In many cases, the cells already show an inconspicuous behaviour after a few minutes (hazard level ≤ 2).Cycling of damaged cells indicates that the risk of ISC is still high due to irreversible structural changes.This statement is supported by a spontaneous TR during cycling in the CV phase of the 255 th full cycle.
The risk analysis of damaged cells by using non-destructive electrochemical methods (selfdischarge, ICA, pulse) is not reliably possible.For timely fault detection, it is necessary to detect the first voltage drops by high-frequency sampling and to exclude further operation of the conspicuous cell.

Figure 1 .
Figure 1.Different voltage characteristics of an ISC compared to the open-circuit voltage U OCV of the undamaged cell, classified here as different ISC types.Adapted from [34].

Figure 2 .
Figure 2. Placement and nomenclature of temperature measurement points on the 10 Ah cell with respect to position of penetration.Detailed description and type of sensor are listed in the corresponding table.

T − 9
* * * and T − 9 ′ , T − 9 ′′ , T − 9 ′′′ ) are located in a distance of 7 mm, 14 mm and 21 mm from the penetration point.On the bottom side of the cell (negative z-direction) exactly opposite the puncture position is the temperature measuring point T − 9 − Bottom.The temperature T − 9 − Bottom is an indicator for the heat conduction through the individual cell layers.The other temperature measuring points T − 3 to T − 15 are used to determine the temperature of the overall cell.At these positions, marked in navy blue in Figure 2, 1.5 mm thick sheath thermocouples (manufacturer: TC Direct; type: 405-008) which are integrated into the upper bracing plate are used to measure the temperature.Based on all temperature sensors mentioned previously, the mean cell temperature (T − Mean) is calculated as an unweighted average (see Eq. (1)).

Figure 3 .
Figure 3. Tip of needle with integrated thermocouple on the left side and an additional zoom view of the foremost tip on the right side.

Figure 4 .
Figure 4. Schematic experimental set-up for the generation of a local, single-layer ISC.Different penetration levels marked with red arrows (see also Sec. 3.1).Information on material thickness taken from [92,102].

Figure 5 .
Figure 5. Test rig for controlled needle penetration consisting of the upper part with actuators and sensors (orange), as well as the test chamber below (red), with air inlet (blue) and smoke extraction system (green).

Figure 6 .
Figure 6.General flowchart for the generation and characterisation of an ISC.

( a )
Development during first the 5 s.(b) Development during the initial 650 s.(c) Development over total test time.

Figure 7 .
Figure 7.Typical ISC characteristic triggered by precise needle penetration.At the beginning the SOC of this cell 1.6 is 100 % and penetration is stopped after the 1 st ISC (penetration depth ∆X = 0 µm)

Figure 8 .
Figure 8. Dynamic ISC characteristic (type IIb) of cell 2.4, which is penetrated by an additional 25 µm after the 1 st ISC at a SOC of 100 %.Abrupt temperature increase develops due to triggered exothermic side reactions.

Figure 9 .
Figure 9. Voltage drop (top) and released joule heat (bottom) of cells 3.3 and 3.6 immediately before the beginning of the TR.

Figure 10 .
Figure 10.Replication of dynamic ISC behaviour using variable parallel-connected resistors.Currentconducting paths shown in red as well as a currentless path in grey.

Figure 12 .
Figure 12. a. Capacity curve and b. self-discharge of a reference cell and three damaged cells 1.6, 2.1 and 0.1.The cyclic load was 1C (CCCV) at 25 • C.

Figure 13 .
Figure 13.Characteristic of cell 0.1 before and during the TR.a. longer time range and b. detailed view during the occurring TR.

Figure 14 .
Figure 14.a.Position of the corresponding images above and b. the images of the damaged cell layers (A -G) and the molten penetration needle (N) of cell 1.6.a. Classification of position and component in the cell 1.6

Figure 16 .
Figure 16.General melting characteristics of the separator (a.) and the needle tip (b.) as a function of the maximum temperature T − 9 * max .Twelve cells are considered here in which no significant exothermic side reactions occurred.

Figure 17 .Figure 18 .
Figure 17.a. Position of the corresponding images above and b. the images of the damaged cell layers (A -G) and the molten penetration needle (N) of cell 2.4 with a Partial Thermal Runaway.a. Classification of position and component in the cell 2.4

Table 2 .
Selected datasheet properties of the SLPB98106100 pouch cell from Kokam.

Table 3 .
Overview of recorded data, with optional information regarding the accuracy and the sampling rate.

Table 4 .
General cell numbering for different penetration depths ∆X.

Table 5 .
Summary of the evaluation features to characterise the ISC behaviour.R min.ISC calculated by Formula (2).

Table 7 .
Listing of the key features according to Tab. 5 of six needle penetrations with the experimental conditions: 100 % SOC and a penetration depth ∆X of 0 µm.

Table 8 .
Listing of key features according to Tab. 5 of six needle penetrations with the experimental conditions: 100 % SOC and a penetration depth ∆X of 25 µm.

Table 9 .
Listing of the key features of six needle penetrations with the experimental conditions: 100 % SOC and a penetration depth ∆X of 100 µm.

Table 10 .
Overview of the performed cell-specific post-characterisations.
Cell from pre-test penetration depth first 25 µm and ≈500 s later another 75 µm † † Development of current over self-discharge duration (I ISC (t start ) − I ISC (t end )).